Supplementary materials for Is there a local knowledge advantage in federations? Evidence from a natural experiment André Schultz and Alexander Libman May 20141

1. The Russian forest fires of 2010 ............................................................................................ 2 2. The local origin index ............................................................................................................ 8 3. The federal connection dummy ............................................................................................ 12 4. Model assumptions ............................................................................................................... 14 5. Quality of bureaucracy and patriotism of elites ................................................................... 16 6. Extreme bounds analysis ...................................................................................................... 26 7. Vice-governors ..................................................................................................................... 28 8. Case study: Ryazan region ................................................................................................... 30 9. Governors` tenure and regional residence ............................................................................ 33 10. Governors’ talents and experience ..................................................................................... 36 11. Forest fire data in Russia .................................................................................................... 42 12. Other robustness tests ......................................................................................................... 44 13. Definitions of the variables ................................................................................................ 48 14. References .......................................................................................................................... 53

1

The supplementary materials were published on the personal website of Alexander Libman.

1. The Russian forest fires of 2010 Since gubernatorial appointments and tenure duration can be endogenous to the economic performance of the regions, this paper attempts to identify local knowledge effects by using the forest fires of 2010 as an exogenous source of variation across Russian regions. Using weather shocks as a natural experiment has become common practice in econometrics in the recent years (Miguel et al. 2004; Bracanti 2007; Baes and Santos 2007; Boettke et al. 2007; Healy and Malhora 2010; Burke and Leigh 2010; Brückner and Ciccone 2011). Before using our case as external shock we have to ensure that a number of assumptions hold. First, it is necessary to show that the allocation of forest fires was indeed random and, what is more important, uncorrelated with the past experience of the forest fires in Russia and hence could not have been predicted by the governors (by taking precautionary measures) or by the federal government (by appointing certain types of governors to regions with higher risk of forest fires). Alternatively, the wildfires can be considered exogenous if the fire magnitude substantially exceeded previous incidents, so that preparing in advance was impossible again. Second, the impact of the forest fires on the national economy should be substantial enough to be considered by the governors as a crucial challenge requiring a lot of effort (otherwise the outcome variation could be simply determined by a lack of attention, which would mean a lot of noise in our data). In what follows, we will systematically demonstrate that the forest fire were unpredictable and uncorrelated with previous wildfire experience and their damage was extraordinary large. 1.1 Temperature anomalies “We can say that for the last 1000 years we have not seen anything comparable in terms of heat.”Aleksandr Frolov, chairman of the Federal Service for Hydrometeorology and Environmental Monitoring (RIA Novosti, 2010, August 9th) The forest fires of 2010 were caused primarily by two anomalous factors: an unprecedented heat wave with record high temperatures (as part of anomalous atmospheric conditions in the northern hemisphere in general) and strong winds during the summer (in our regressions, we control for various proxies of temperature and temperature anomalies, as well as wind velocities, as discussed below). For the summer months in 2010 (June, July, and August) the Russian Federal Service for Hydrometeorology and Environmental Monitoring measured a countrywide average temperature anomaly of 1.81 degrees Celsius2 which made it the hottest summer since official weather records began in 1936 (Russian Institute of Global Climate and Ecology 2010b). A weather station in the Southern European part of Russia (Kalmykia) measured 45.4 degrees (July 12, 2010), which is the highest temperature ever recorded on Russian soil and breaking the previous record of 1940. At the same time, the spring weather did not herald the approaching historic summer. In the spring months (March-May) the countrywide average temperature anomaly amounted to only 0.71 degrees (Russian Institute of Global Climate and Ecology 2010a).3 While the national average temperature was high, it conceals that the different regions reported positive and negative temperature anomalies (from minus 4 up to plus 6 degrees). As illustrated by Figures 1-3 the heat wave did not hit every region with the same severity. While the Ural regions were largely spared from temperature anomalies, some Siberian regions even 2

Throughout the paper temperatures will be specified in degrees Celsius. Temperature anomalies are defined as deviations from the average temperature of the reference period (1961-1990). 3 Considering temperature anomalies during the spring months (March-May) since 1936, the spring anomaly of 2010 ranks only 22nd (reference period 1961-1990).

experienced negative temperature anomalies. In contrast, the Central, Southern, and Far Eastern regions measured very high positive temperature anomalies. Moreover, the temperatures anomalies of 2010 were only moderately correlated with the long term temperatures in individual regions (the correlation coefficient of 0.16). This provides a strong argument in favor of using our empirical setting, as the presence of forest fires (or lack thereof) is determined precisely by the extent of temperature anomalies (which make the existing precautionary measures insufficient). It is safe to say that for each Russian region the degree of exposure to the heat waves and winds was entirely unpredictable and not driven by persistent repeating weather conditions. 1.2 Forest fires “…this (the forest fires) is a real natural disaster which happens once in 30-40 years.” Dimitri Medvedev, president of the Russian Federation (First Channel of the Russian federal TV, 2010, July 31) Although forest fires of course happened in Russia before, the destructive effects in 2010 were substantially larger than usually. Overall the forest fires in 2010 have been regarded one of the most devastating ever recorded. The period of permanent wildfires lasted from the end of July until the beginning of September 2010. According to official data by the Russian Statistical Agency Rosstat in 2010 the amount of burned trees has increased by over three times in comparison to the average in 2000-2009 (although the number of forest fires reported increased only by approximately 23 percent). Wildfires affected 2,026,873 hectares of forest area all over the country (versus 1,404,732 hectares on average in 2000-2009). As a result of the heat, the Russian economy suffered under significant damage, which, according to some estimates, reached the level of 40 bln USD, or 1.4% of GDP, including material damage, human losses (calculation based on statistical life value) and output decrease due to productivity losses as a result of temporary suspension of production in several industries, reduced working hours and drought (Porfiriev 2012).4 As of August 6, more than 100 villages were completely or partially destroyed by fires. The air pollution due to forest fires caused a substantial number of excessive deaths in large cities5 and forced a number of companies and agencies (including many foreign embassies) to stop their current operations in Moscow. What is even more important is that the forest fires indeed followed a path difficult to predict based on previous observations. The actual destructiveness of the forest fires among regions in terms of burned trees in 2010 is uncorrelated with the previous years: the correlation coefficient between the indicator of 2010 and of 2000-2009 is equal to minus 0.03, i.e. the allocation was almost random as opposed to the past experience. As we have already pointed out, the area covered by forest fires increased by roughly two times as opposed to the average in the previous decade; but in some regions the increase was much larger (in the Central and Volga federal districts the area covered by forest fires went up by 42 and 51 times respectively), see Figures 4 and 5. Finally, the entire natural disaster happened within two months (mid July until mid September, depending on the region), partly falling in the national vacation period which made it technically impossible for the regional administration to replace governors as an immediate reaction to their crisis management. Traditionally, regional 4

For example, the draught destroyed about 20% of the country’s wheat crop (Hernandez et al. 2010), forcing the government to impose export restrictions on grain. 5 For example, in July 2010 the mortality rates in the city of Moscow increased by 50.7%, in Vladimir region by 18.4%, in Ivanovo region by 18.3%, in Moscow and Tula regions by 17.3%. In August some reports indicate an increase of mortality from the average 360-380 people per day to 700 people per day. The overall excessive death from the heat and wildfires is estimated at the level of 55,800 people in June and August 2010.

governors are replaced either at the beginning or at the end of the year. This again makes reverse causality between our dependent and explanatory variables almost impossible. 1.3 Forest administration With its approximately 7.8 million km2 Russia has the largest forest area in the world. 45.4% of Russia’s soil is covered by forest. In administrative terms the Russian forest is divided into forest located on land of the forest fund (99.5%) and forest located on other territory (0.5%). The former is property of the Russian federal government, while forests in the other categories are owned by regions, municipalities, cities, the military, private companies etc. Until 2007, the forest territory of the forest fund was managed by the Federal Forestry Agency. After the enactments of the Forest Code in 2007 the management of the forest fund was decentralized to the regions. The willingness for decentralizing forest management in an era of political centralization was primarily driven by the perception that forests are an ‘economic asset’ which needs to be commercialized. The idea was to lease forests to enterprises for economic use and simultaneously transfer the obligation for forest management to the leaseholder. Because the federal center believed that this policy can be implemented most effectively on the regional level it decided to decentralize forest administration to the regional governments. At the same time the federal law prescribed the regions to outsource the forest fires prevention to private parties based on public tenders, where the only selection criterion acceptable was the price. Proof that the participating service providers owned the necessary technical equipment, or had experience in forest fire extinction was, according to the law, not required. Often the cheapest, but not the most qualified firm was awarded with the contract.6 These contractors struggled already with occasionally occurring forest fires, but were desperately overstrained with a natural disaster of the scale occurring in 2010 (see also Blokov 2010).

6

There is rich anecdotal evidence on the companies wining public tenders for forest fire protection without necessary qualifications. For example, in Mordovia region the tender was won by a company with 50 employees and 2 cars (27% of Mordovia region is covered by forests which is equal to 710 thousand hectares).

Figure 1: Land surface temperature anomalies for July 20-27. 2010 compared with the temperatures for the same dates for 2000-2008. Source: NASA Earth Observatory (http://earthobservatory.nasa.gov)

Figure 2: Monthly average temperature maps

June 2010

July 2010

August 2010

Summer 2010: June-August

Source: Russian Institute of Global Climate and Ecology (2010b)

Figure 3: Monthly average temperature probability map The maps below show the spatial distribution of 263 weather stations which measure daily climate conditions throughout Russia. The color of each weather station corresponds to the percentile in which the measure temperature falls relative to the temperatures measured for the respective period since 1936. Hence the redcolored weather station indicated that 90-100% of the past temperatures recorded by this weather station for the respective period in the year fall below the temperature recoded in 2010.

June 2010

July 2010

August 2010

Summer 2010: June-August

Source: Russian Institute of Global Climate and Ecology (2010b)

Figure 4: Satellite forest fire monitoring (29.07.2010)

Source: “Kosmosnimki forest fire monitoring” (http://fires.kosmosnimki.ru/) Figure 5: Satellite forest fire monitoring (31.08.2010)

Source: “Kosmosnimki forest fire monitoring” (http://fires.kosmosnimki.ru/)

2. The local origin index One of the most influential reforms on Putin`s centralization agenda was the replacements of gubernatorial elections by federal appointments in 2004.7 A direct consequence of the federal selection of regional governors was the appointment of governors, who came from territories thousands of kilometers away from the regions they have to rule and who fundamentally have never lived in their region of office before.8 While in some region the federal center appointed outsider governors, in other regions it chose candidates from the local political elite. We use the variation of local origin among Russian governors to study the effect of local knowledge on the forest fire management of governors. 2.1 Method The local origin index measures the relative period a governor has spent in his region of office before inauguration. For this purpose we determine the governor’s duration of residency and calculate the age at taking office. The ratio of the length of stay to the inauguration age yields the degree of local origin. Essential for this kind of calculation is an extensive analysis of biographical information data. We have utilized the data from publicly available biographies of all Russian governors and extracted information on the place of birth (also serves as a proxy for the region of adolescence, if no other information is available), institution of higher education and professional career path. For the three stages of life we determined the date and geographical region (in terms of the 83 Russian regions). These three phases of life contribute to the degree of local origin and can be related to different aspects of local knowledge. Thus, in the childhood and adolescence years one encounters moral beliefs and traditions. In the education phase one is able to set up and expand a network of regional contacts, while during the professional career one is confronted with specific regional problems, as well as advances the contact network. We are aware of the fact that by assessing a relative and pre-inauguration local origin score we face some problems: A relatively young governor who has spent his entire life in his region of office might receive the same local origin score, as someone who has “crowned” his political career with gubernatorial office. Furthermore, a governor who has been appointed recently may behave differently as a governor who has already 20 years of gubernatorial experience. However, we want to differentiate different sources of local knowledge by looking separately on local origin, tenure, and age. 2.2 Assumptions When analyzing the governor’s pre-inauguration biographies and calculation the length of residence in the region of office we make a number of assumptions and corrections. First, we are aware of the fact that only looking at the place of birth as a proxy of childhood and adolescence is not sufficient. Since it is possible that a governor might have moved to another region within this first phase of life, we check for any inter-regional mobility within the first 18 years and when necessary correct our estimations accordingly. Second, we assume a period of five year of education when the governor has studied in university. This assumption 7 The system of appointments was abolished in 2012 following the mass protests against the falsified Duma elections and replaced by highly regulated elections of governors. However, during the period of investigation of this paper, appointment system was fully in effect, and there was no evidence of the system being changed in any foreseeable future. On centralization in Russia under Putin see Idrisova and Freinkman (2010) and Ross (2010); for gubernatorial appointments see Turovskii (2010). 8 To provide an example, Aleksei Gordeev, governor of Voronezh region appointed in 2009, was born in Frankfurt-an-der-Oder (German Democratic Republic), spent his childhood in Magadan about 12,000 kilometers away from Voronezh, studied in Moscow (“only” about 600 kilometers away) and continued his political career in the Moscow region and afterwards in the federal ministry of agriculture.

is unavoidable, as most biographies indicate only the year of graduation and omit the year of enrollment. Five-year period corresponds rather well to the standard duration of the university studies according to the Russian educational standards until recently. We have adopted this assumption for governors who have only completed secondary education, studied on a parttime basis, or prolonged their education by a subsequent PhD degree. However, we consider only the places of higher education which were completed. Third, we neglect military service in our calculations. All governors, regardless whether during Soviet or post-Soviet times were required to serve 2 years in the armed forces. Most governors completed their military service either before, or directly after their university studies. However, most biographies do not specify in which regions the governors were stationed (this information is classified according to the Russian law). Furthermore, it is unlikely that soldiers in Russian army had any chance to socialize with people of the region they serve in. We adopt this assumption for governors who made their professional career in the army. Fourth, in cases were the governor has made a career as businessmen his place of residence is not always specified. For these cases we assume the headquarters of the company to be place of residence. For small regional companies this assumption is unproblematic as locality of business operations and management coincide. For large inter-regional companies (which headquarter is always in Moscow city) we made a justified judgment, based on further available information. 2.3 Typology A governor who has spent over 70% of his life in his region of office is considered to have a high level of local origin (score 4). On average the governors with a score of 4 have spent 89% of their life in their region of office. In order to fall in this category a governor with the average inauguration age of 53 years should have spent a maximum of 12 years outside his region. If one assumes a 5 years leave in order to study in a distant university (e.g. in Moscow as many governors did) this leaves only room for a maximum 7 years period outside the region (much less for younger governors). A governors who has lived between 20-70% of his lifespan in their region of office is considered to have intermediate level of local origin (score 3). On average the governors with score 3 have spent 44% of their life in their region of office. The last group of governors spends only 20% or less for their life in their region of office. These are governors who made a rapid career in the region (within a few years although being outsiders) and external politicians who have been appointed by Putin or Medvedev. On average the governors with a score of 1 and 2 have spent 3% of their life in the region of their office. To account for the heterogeneity of the Russian Federation the last group is additionally subdivided between governors who originate from close by regions (score 2) and governors from distant region (score 1). As a measure of distance we use the federal district classification. Federal districts are groups of geographically close regions, which have been established by Putin in 2000. If the governor originates from a region within the same federal district he receives a score of 2 if he originates from a region of a different federal district he receives a score of 1. 2.4 Example For illustrative reasons we provide an example of a record which presents the necessary and used biographical data in order to calculate the local origin index. Similar records have been constructed for all Russian governors. The data is presented for the specific of example of Belgorod region, in the Western part of Russian (Central federal district). Notice, that geographical units which are specified in brackets refer to the respective region (administrative names such as republic, krai, or okurg are omitted). The governor of Belgorod region, Yevgeny Savchenko, was born, raised, and educated in Belgorod region (primary and

vocational education). After he completed his higher education in Moscow City, he moved back to Belgorod where he worked until the dissolution of the Soviet Union. Before he was appointed by Yeltsin he spend three years in the in the ministry of agriculture and production of the Russian federation. In total we estimate that he spend 35 of his 43 pre-inauguration years in Belgorod (81%) which qualify him for the highest local origin score of 4. Region: Belgorod Governor: Yevgeny Savchenko In office since 1993 Date of birth: 1950 (Belgorod) Education: Moscow academy of agriculture, 1976 (Moscow city) Career stages: 1976-1990, collective farm, state farm, district administration (Belgorod) 1990-1993, ministry of agriculture and production (Moscow city) Inauguration age: 43 Years spend in his region: 35 Years spend outside his region: 8 Share of years spend in the region: 81% Local origin index: 4 Local origin dummy: 1 Federal connection dummy: 0 Figure 6: Distribution of regions according to the value of the local origin variable

0

10

Number of regions 20 30

40

Local origin index

1

2

3

4

2.5 Examples of governors in each sub-group of the local origin index In order to illustrate the different degrees of the local origin index we will provide four explanatory examples of the governors in power during the period of our investigation. •

The mayor of Moscow City, Yuri Luzhkov was born, raised and educated in Moscow. After graduating from the State University of Oil and Gas in Moscow he commenced his professional career in the chemical industry, as well as his political career in the

•

•

•

administration of Moscow city until he eventually became mayor in 1992. Thus, this governor unambiguously received the highest possible score equal to 4. The governor of the Smolensk region, Sergey Antufyev, was born in Kazakhstan and studied at the Kazan State Technical University in the Tatarstan region. After his graduation he moved to Smolensk where he commenced his professional and political career in the city administration before he was appointed governor of Smolensk in 2007. In this case the governor’s local origin score is 3. The governor of the Ivanovo region, Michael Men, was born in the Moscow region and studied at the State University of Oil and Gas in Moscow. After graduation he pursued his professional and political career in the administrations of Moscow region and Moscow where he eventually reached the position of vice-mayor in 2002. However, in 2005 he was appointed governor of Ivanovo region. Since Ivanovo is geographically not very far away from Moscow and correspondingly falls within the Central federal district Men’s score accounts to 2. The governor of Kamchatka region, Aleksey Kuzminzkiy was born in Kemerovo (Siberian Federal District) studied in St. Petersburg (Northern Federal District) and worked in Moscow (Central Federal District). In 2005 he was appointed vice-governor and in 2007 governor of the Kamchatka region (Far Eastern Federal District). Since in this case the governor’s place of origin and work was geographically extremely distant from the region to which he was appointed he falls within the local origin 1 sub-group.9

2.6 Alternative measures of local origin We compile a number of alternative measures of local origin including a binary local origin dummy and two continues local origin measures: i) years of regional residency until inauguration and ii) years of regional residency until forest fires (thus including tenure duration). All variables are highly correlated and generate the same empirical results. For example the correlation between the local origin and the two continuous local origin measures equals 0.92 and 0.91 respectively.

9

The way in which the local origin index was constructed could be debated. Particularly, the difference between 1 and 2 may be superficial. The borders of federal districts do not necessarily represent a proper division of the Russian territory in any way. Moreover regions within federal districts can be very heterogeneous and may be located far away from each other (e.g. Siberia or Far East). In the same way, the difference between 3 and 4 is in some cases in flux. Therefore, we checked for the robustness of our findings by using a number of alternative proxies, entirely confirming our results.

3. The federal connection dummy Informal connections in non-democracies are difficult to uncover. While media reports and academic studies are full with references to “clans” and “networks” in the Russian elite (Wedel 2003), they rarely provide convincing empirical data. However, the role of informal networks in Russia cannot be underestimated.10 To capture federal connection we create a dummy which is one whenever a governor has worked in federal institutions since 2000 and before his inauguration (and zero otherwise).11 Typically, the governors with federal connections have worked in ministries, the administration of the president (an institution providing direct support to the president) and as plenipotentiaries of the president, i.e. the appointed representatives of the president to various groups of regions responsible for supervising the governors on behalf of the presidential administration. The latter group is officially part of the presidential administration. In Russia all federal agencies operate in the regions through their own branches rather than by delegating authorities to the regional governments. We count the positions in regional branches of federal agencies as ‘federal connections’ as well. The positions as members of the parliament do not count as federal connections, since in this case the links to the federal bureaucrats (who are actually in charge of distributing resources) are not necessarily present (the Russian parliament is essentially powerless and accepts almost all suggestions of the executive without further debate). 3.1 Examples of governors with federal connection •

•

The head of the Altai Krai, Alexander Karlin, was appointed governor in 2005, having worked before in the ministry of justice (2000-2004) and the administration of the president (2004-2005). In 2002 he was appointed first deputy minister of justice, i.e., the highest rank below the position of the minister himself. The governor of Orel region since 2009, Alexander Kozlov, worked as a deputy head of the presidential administration (1999-2004) and deputy minister of agriculture (20042009).

3.2 Examples of governors with both local origin and federal connections and without both •

•

10

The head of Bashkortostan, Rustem Khamitov, spent the longest period of his life in Bashkortostan,12 after which he left the region for 8 years, working in the federal tax agency and the federal water resource agency, before returning as governor to Bashkortostan in 2010. Thus, this governor has both local origin and federal connections. Vyacheslav Nagovitsin, head of Buryatia, spent his entire career in Tomsk, where he worked in the regional administration (since 1999 as vice-governor) before moving to Buryatia. This governor has neither local origin nor federal connections.

For its role in the Russian judiciary see Schultz et al. (2014). The previous literature has measured federal connections, for example, if the region and the federal center were governed by the same political party (Grossman 1994; Khemani 2004; Soares and Neiva 2011); but one cannot use this approach in a non-democracy like Russia, where almost all governors are members of the proPutin United Russia party. 12 Rustem Khamitov was born in Kemerovo region (Siberia). However, as a child he moved to Bashkortostan (Volga). Later, he left the region for Moscow City in the 1970s where he graduated from prestigious state technical university (Bauman Technical Institute). 11

3.3. What does federal connection dummy measure? One could hypothesize that appointment of federal officials as regional governors constitutes a form of exile for those who have lost their influence on the federal level. However, this interpretation is inconsistent with the growing number of governors with federal connections since the gubernatorial elections were abolished. While in 2007 only 6 governors had close ties to federal institutions, by 2010 already 17 regions were headed by politicians with connections to Moscow. By 2011 almost every newly appointed governor has federal connections. Since during this period the Russian federal elite exhibited very high continuity and there is little evidence of strong and open conflicts leading to ‘federal elite cleansing’, it is unlikely that these positions were filled with exiled bureaucrats – on the other hand, the use of appointments to increase control over regions as part of the overall centralization trend in the Russian federalism seems to be more likely. 3.4. The potential effects of federal connections Similarly to local origin, federal connections dummy may have heterogeneous effect on performance of the governors. Federal connections could improve the performance of the governor due to the better access to the federal administration. Furthermore, politicians with federal connections may consider a reappointment at a higher level of political hierarchy more likely and therefore attempt to perform better to advance to higher stages of career ladder. In China, for example, the performance-based promotion system of regional bureaucrats is perceived to be one of the major advantages of the existing political system. Alternatively, federal connections could also let the governor consider his appointment a merely short-term assignment, thus reducing the effort and focusing on rent-seeking. Ultimately, the effect depends on the promotion system for bureaucrats in Russia; and since, as mentioned, Russian governors’ promotion typically is unrelated to economic performance of their regions, large rent-seeking by federally connected governors is probable (Libman et al. 2012). However, in case of forest fires of 2010, due to enormous attention paid by the federal government to the performance of governors, again, we should expect federally connected governors to perform better and to utilize their connections to the federal administration to the full extent. 3.5. ‘Federal knowledge’? Throughout the paper, we present our key explanatory variables as ‘federal connections’ and ‘local knowledge’. However, the ‘local knowledge’ we describe, i.e., knowledge of regional elites and bureaucracies, is also derived from connections governors have within the region (Binz-Scharf et al. 2012). Thus, it would be possible to refer to ‘federal knowledge’? The reason why we do not do that is the context in which ‘local knowledge’ is referred to in the literature. ‘Local knowledge’ is not just any knowledge of the region – it is the knowledge derived from observations and “casual empiricism or thoughtful speculation” (Fischer 2000: 195) rather than formal studies. Thus, ‘local knowledge’ has as its counterpart in formal, scientific knowledge. The argument of the federalism literature is that precisely the fact that the local politicians have this informal, observational ‘local knowledge’ improves their performance. This is what we tried to test in the paper. Federal connections would also generate knowledge – but, again, it would be knowledge of informal, observational nature. Thus, by referring to ‘federal knowledge’ as a counterpart of ‘local knowledge’ we would make it more difficult to fit our argument in the language of the theory. Therefore, we prefer using the term ‘federal connections’.

4. Model assumptions 4.1 Dependent variable We compile a dependent variable which equals to the ratio of forest area covered by fires to the number of reported forest fires in 2010. The dependent variable is highly skewed to the left and therefore requires a logarithmic transformation (see Figure 7). However, it includes values equal to zero or less than one. Simply taking logs would lead to two problems. First, we would obtain negative values for some observations which is meaningless in our context. Second, it would exacerbate the impact of outliers if the original dependent variable was less than one. To prevent this problem, we apply the standard scaled OLS approach (e.g., which is also used in the international migration literature). We simply add one to each value before calculating the natural log of our dependent variable. Although the data on forest fires and on regional economic and political characteristics is available for at least a decade, we stick to a cross-section analysis. Using any other year except 2010 we not only lose the advantage of absent reverse causality, but also have to deal with the fact that in most region previous forest fires were rather limited in scope and therefore received little attention from governors We also run regressions with forest area covered by fire as dependent variable while simultaneously controlling for the number of forest fires. For our results this does not make a difference.

0

0

5.0e-04

.1

.001

Density

Density .0015

.2

.002

.3

.0025

Figure 7: Distribution of the independent variable

0

1000 firearea_fire2010

2000

3000

Distribution of the ratio of forest area covered by fire and reported forest fires

0

2

4 logfirearea_fire2010

6

8

Distribution of the log ratio of forest area covered by fire and reported forest fires (after transformation)

4.2 Urban population All main regressions include urban population (log) in the set of covariates.13 Urban population is neither correlated with the local origin index nor with the federal connection dummy. The respective Spearman and Pearson correlation coefficients are below 0.1 and insignificant. In the same way, if one computes the mean comparison, there is no significant difference in the urban population between regions with different level of local origin or federal connections. Thus, we can use the interaction terms without encountering the problem of multicollinearity. 13

Similar to other region-specific covariates (e.g., regional forestry expenditures), we average urban population over 10 years (2000-209) to capture the regional long-term characteristics.

We use urban population to control for the ‘human factor’ in forest fire emergence and expansion. Moreover, urban population serves as control for the varying degree of federal attention. We argue that more populous regions are under the strict scrutiny of the federal center making information transmission by the governors less relevant (e.g., about the threat by forest fires).14 However, for smaller regions which receive less attention by the federal center, the role of the regional governor as channel for information transmission between the federal and regional level is particularly important. Hence, we expect the effects of local origin and federal connections to be significant only for regions with small and medium level of urban population. An alternative interpretation of the role of urban population would claim that the federal government generally disregards small regions and provides assistance only to larger regions. If this were the case, however, we would not expect to see any significant effect of governorspecific characteristics for forest fire management. The ‘one-man control approach’, which we described in the main part of the paper, also applies to the federal level. The Russian president and members of the cabinet personally oversee forest fires fighting in particularly important regions and check most decision of the bureaucrats.15 The more pronounced the federal intervention is, the less important the role of governors becomes. Thus, if this alternative conjecture would be true, we wound find generally better performance of large regions in combating forest fires, but no variation in the performance of large regions depending on the personality of the governor. 4.3 Composition of the sample Our dataset in the basic specification includes 71 regions. We have to exclude a number of regions from our sample. First of all, we exclude Chechnya and three and the autonomous okrugs. While for war-torn Chechnya no reliable data exists, the autonomous okrugs are administrative sub-units of other regions what makes it difficult to disentangle the explicit line of authority. Second, we exclude Moscow City and St. Petersburg City because they have no forests (the park areas in the cities are outside the scope of our analysis). Third, we exclude Chuvashia region, because the governor of Chuvashia resigned in August 2010 (in the midst of wildfires) after completing his fourth term in office. There is no evidence that this decision was influenced by any events connected to the forest fires.16 Nevertheless, for this region we cannot confidently isolate the effect of the resigning governor from this successor (although occasionally wildfires happen throughout the year, the disastrous forest fires in 2010 were mainly concentrated in end of July until the beginning of September, which is therefore the focus period for our investigation). Finally, we exclude five regions (Tula, Kalmykia, Ingushetia, Kabardino-Balkaria, and Northern Ossetia) which reported no forest fires in 2010 (and hence the dependent variable described above is meaningless). We assume that these regions were not affected “by chance” or simply as a result of their topography (e.g. mountainous area with little forest coverage in the North Caucasus). We will relax most of these assumptions in the robustness checks

14

We acknowledge that urban population is not the only possible proxy for federal attention and therefore use other proxies in the robustness checks.. 15 E.g., Russian president personally commented on the decision of the mayor of Moscow – the most populous region of Russia – to interrupt his vacation during forest fires, see RIA Novosti, 2010, 10 August. 16 The governor of Chuvashia was in power since 1994, with his last term starting in August 2005 and expiring in August 2010; thus, the end of his term simply happened to coincide with forest fires disaster. After the governor stepped down from office he continued his political career and was eventually appointed federal minister of agriculture.

5. Quality of bureaucracy and patriotism of elites Although our discussion shows that governors play an important role in the forest fire management, it does not necessarily make their influence decisive. The empirical case we investigate removes the problem of reverse causality, but we still have to deal with the problem of omitted variable bias. Specifically, we have to discuss other region- and governorspecific characteristics, which potentially influence the ability of regions to combat forest fires and could be correlated with governors’ local origin or federal connections. In this section we focus on two sets of variables.17 First, some region may have better overall quality of administration and forest governance. If the personal properties of governors are correlated with these characteristics, we would encounter a spurious correlation problem. Second we have to look at the intrinsic motivation and power of local elites. Highly patriotic and powerful elites could be ready to engage into protection of their region, on the one hand, and ensuring that only a local governor gets an appointment by pressuring the federal administration, on the other. In this case the elites could perform excellently while combating forest fires and at the same time be the reason for the local origin of the governor ruling in the region. In what follows, we look at whether these variables are correlated with local origin and federal connections. Specifically, we regress the governors’ characteristics on these variables and perform an F-test on the joint significance. If the test is insignificant, the governors’ characteristics are orthogonal to state capacity and patriotism of local elites. It is also possible that the correlation between governors’ characteristics and state capacity is significant, but goes in the opposite direction to that suggested by our results. For example, high local origin could be correlated with low state capacity, but at the same time with better performance in forest fire management. It means that the positive effect of local knowledge is powerful enough to offset the negative effect of low state capacity, and we again do not need to be concerned about our results. On the contrary, they provide only conservative lower bound of local origin effects.18 5.1 Quality of bureaucracy There exists a rich literature measuring state capacity (Steinmo 1993; Cheibub 1998; Hendrix 2010; Hanson and Sigman 2013). Most of the studies, however, focus on the state capacity of independent states. These proxies are not necessarily applicable for measuring capacity of sub-national governments. Still, in what follows we will use a very broad set of indicators potentially capturing state capacity. First, we control for the number of public servants; the share of public servants with university education; and the share of public servants participating in additional professional 17

In addition, we could be concerned about the role of local interest groups, which are discussed in greater detail in the discussion section of the paper. See also Prud`homme 1995; Goldsmidt 1999; Tanzi 2000; Treisman 2000; Fishman and Gatti 2002; Arikan 2004; Reinikka and Svensson 2004; Slinko et al. 2005; Gurgur and Shah 2005; Persson and Zhuravskaya 2011) 18 We run OLS regressions, but also replicate results using ordered logit and logit for local origin and federal connections respectively.

training.19 In addition, we also look at the number of independent public agencies in the regional government. Larger number of agencies could make coordination problems more pronounced, but also could allow for greater specialization and thus better performance of individual bureaucrats. We also run a regression where we control for the number of municipalities in the region. More municipal governments could make the task of coordinating forest fire fighting activity more challenging. Second, we add a number of aggregated indices directly attempting to measure the quality of bureaucracy and economic institutions in the regions. Specifically, we control for three indicators developed by Aleskerov et al. (2006) measuring the quality of regional bureaucrats using an analysis of their operations and practices; two indicators published by the Ministry of Regional Development and based on the aggregation of multiple regional economic and social characteristics measuring the performance of regional governments and the improvement of their performance since 2007; two corruption indices based on a large survey implemented by FOM (one of the leading public opinion agency) measuring experienced corruption and perception of corruption; and an index of sub-national democracy developed by Moscow Carnegie Center (the leading think tank in this area) based on an expert survey.20 In an alternative specification we substitute the index of democracy by the index of the freedom of the press.21 Third, we use a number of proxies capturing the capacity of government, which are broadly used in the literature, but are unfortunately not necessarily applicable for the Russian case. Specifically, they look at the extractive capacity and the ability of the government to combat crime. For Russia the problem is the following: federal, regional and local taxes are collected exclusively by the federal tax agency, while all police functions are exercised by the federal Ministry of Interior. Thus, by looking at these indicators, we do not measure the performance of regional bureaucrats (who, as mentioned, are responsible for forest fire fighting in Russia), but rather the performance of regional branches of federal agencies. Since the massive forest fires also involved a number of federal agencies, looking at these variables still may be warranted. Therefore, we use the following proxies: share of total tax revenue collected on the regional territory in the regional GDP; share of tax arrears (taxes claimed by the government but not paid by taxpayers) in the regional GDP;22 ‘relative extractive capacity’, following Kugler and Arbetman (1997); number of crimes investigated by police per capita. We also add the indicator of infant mortality (this indicator may be less problematic as regions are in charge of a large sector of healthcare system). We also replace total tax revenue by income tax revenue in a further specification, following Lieberman (2002). Fourth, we look at proxies of state capacity specific to the forest sector, which are particularly important for us given the topic of our investigation. First, we use a set of indicators developed by the WWF (one of the recognized international environmental NGOs) and measuring the quality of forest governance in Russian regions in the following areas: organization of forest governance (number of bureaucrats, their resources and quality of their work); legal framework and enforcement of legal provisions; economic efficiency and environmental quality of governance. Second, while some regions in Russia created a 19

A number of papers show that the size of bureaucracy has a significant effect on economic performance of Russian regions, although the direction of this effect differs in individual studies (Dininio and Orttung 2005; Brown et al. 2009; Libman 2012). For a similar approach for Columbia see Acemoglu et al. (2014). 20 Keefer et al. (2011) discuss why democracy should be associated with better disaster management. 21 Besley and Burgess (2002) demonstrate that this variable can have a particular impact on governmental responses to disasters. 22 Treisman (2003), Ponomareva and Zhuravskaya (2004) and Libman and Feld (2013) use this indicator as a proxy for tax evasion.

specialized forest agency, others delegated forest governance functions to more general environmental protection agencies. It is plausible to hypothesize that the regions with specialized forest agencies devote greater attention to forest governance, and thus we control for a respective dummy. We also include the yearly average forest area covered by fire for the period 1992-2009, which may capture the experience of the generally immobile regional bureaucrats with forest fires. In addition, we also run a regression using all proxies described above together. We treat the results with caution, since the number of observations is low and we use numerous controls which may be subject to multicollinearity. Still, this regression allows us to test whether all state capacity variables are jointly significant. Moreover, in all regressions, we also control for log GDP per capita. We use this control primarily to isolate the effect of state capacity from the general wealth of the region. 23. The results reported in Table 1 are unequivocal. The local origin variable is orthogonal to almost all variables measuring capacity of the regional government. The only exceptions are the share of bureaucrats participating in professional training, the level of democracy and the income tax revenues. All three variables, however, are lower in regions run by local governors, indicating that these regions should have lower capacity to deal with crises. The negative correlation with democracy may reflect one of two outcomes. On the one hand, a non-democratic federal government may be more likely to replace leaders in more democratic regions to establish sufficient control over them. On the other hand, leaders of non-democratic regions may be more difficult to replace. Their successors would face resistance of the wellorganized political machines of the old governors and thus unable to achieve the goals of the federal government. The low income tax revenue may reflect stronger resistance of established business groups in these regions to the pressures of federal tax administration. In Russia, most of the income taxes are paid on wages, which are collected by employers. In the 1990s these taxes were extremely low due to massive tax evasion (most employees received their wages in cash without paying any taxes). In the 2000s, the federal government put a lot of pressure on business to improve tax compliance. We could hypothesize that in regions with local origin some companies still have good connections with the regional governors and escape this pressure. Finally, lower participation of bureaucrats in training measures may reflect greater emphasis on personal connections than on formal criteria in regions run by local governors. The results for federal connections are reported in Table 2. Federal connections are associated with lower GDP per capita and lower economic efficiency of forest management. It may indicate that the federal government is more inclined to send its appointees to problematic regions, hoping that they will improve the economic situation and the efficiency of forest management. Again, in both cases it looks like these regions should have lower state capacity. 5.2 Power and patriotism of local elites To deal with the intrinsic motivation of elites and their power, we focus on four sets of variables. The first looks at ethnic heterogeneity of the Russian Federation. It is plausible to hypothesize that in ethnic regions the elites are, on the one hand, more powerful and thus able

23

Some papers suggest that GDP per capita can be used as a proxy for state capacity as well, which is, however, highly debatable. See the discussion in Hendrix (2010).

to determine the appointment of the governor, and, on the other hand, have stronger intrinsic motivation to support the development of their region.24 Thus, we look at the following first five variables: dummy for ethnic republics; dummy for ethnic republics of the Northern Caucasus (these regions have a de facto special status in modern Russia due to the spillover effects of the Chechen war); share of ethnic Russian population; index of ethnic fractionalization and (e) polarization index for the ethnic structure. Second, patriotism (and also power vis-à-vis the federal center) may be correlated with the distance between the regional capitals and Moscow. Arguably, elites and the population of more distant regions may have more specific preferences or may be more likely to expect to spend their entire life in their region and are therefore more ‘patriotic’. Third, we look at the oil-extracting regions, which are likely to have a stronger bargaining position in dealing with Moscow. Fourth, the power of local elites can be also captured by the typical patterns of gubernatorial appointments. Thus we look at tenure of the predecessor governor (as a proxy for the tradition of long-term governors in the region) and number of predecessor governors (as a proxy for elite continuity). The findings for the local origin index, which are reported in Table 3, are again straightforward. The only variable correlated with the local origins of governors is the number of predecessors. Local governors are more likely to run regions, where the change of governors was less frequent in the past. However, this significant result may simply be driven by the fact that the appointment of outsider governors is a more recent phenomenon than the emergence of local governors (who dominated the Russian politics in the 1990s). Indeed, if we control for tenure of the incumbent governor, the number of predecessors variable becomes insignificant. The results for governors with federal connections are more heterogeneous (see Table 4). On the one hand, they are indeed less likely to be found in regions further way from Moscow. It may reflect the strength of local elites, but may also be driven by the self-selection of bureaucrats trying to avoid appointments in ‘distant provinces’. Furthermore, federally connected governors are less frequent appointed in oil-rich regions. Unsurprisingly, governors with federal connections are often present in regions with more frequent turnover of governors in the past. But they also are more often succeeding governors with long tenure. It may, on the contrary, show that the federal administration attempts to appoint governors with federal connections as replacements of powerful and well-entrenched local governors, and thus improve its control over these regions. The last interpretation is in particular corroborated by the fact that federal connections are positively correlated with dummy for the ethnic republics of the Northern Caucasus, where federal government exercises particular effort to ensure its control. 5.3 Further interpretations of the local origin variable Discussing the impact of local elites, we also have to address two alternative interpretations of the local origin variable. The local origin variable could capture the extent of connections of the federal center to insiders in the region. It means that if the center is strongly connected to local politicians (i.e., perceives them as very loyal), it is more inclined to appoint someone among the insiders as a regional governor. However, if this were the case, we should also find particularly high 24

Ethnic regions (which are officially called republics) are regions where particular ethnicities have a special status.

electoral support for Putin and Putin-friendly forces during federal elections in regions run by local governors. Russia is an ‘electoral autocracy’ and ensuring the favorable outcomes for Putin during elections is absolutely essential for the survival of its regional political leaders (Schedler 2009). Simpser (2013) argues that this system require not merely ‘some’ support during elections, but absolute majorities, as a signal of strength. Thus, if the center is particularly well-connected with the regional elites, it would ensure that these elites manipulate voting in the region in favor of the federal government. However, there exists no significant correlation between electoral support the center receives and the local origin index, controlling for the ethnic composition of the region. It would mean that the center does not use its good connections to the regional elites to ensure favorable elections outcomes – which would contradict the available evidence on how the Russian authoritarian state functions. The local origin variable could also capture the impact of local elites on federal appointments. Thus in regions where elites are stronger, they are more likely to ensure that a local politician becomes a governor. The impact of local elites as such, however, does not ensure that the governors demonstrate superior performance as this requires that local elites are strongly intrinsically motivated (i.e. ‘patriotic’). With regards to the patriotism of local elites there exists a substantial variation across Russian regions. In many regions local elites are controlled by large federal business groups (Orttung 2004). In other regions elites concentrate on rent-seeking through utilizing resources remaining from the Soviet period (asset-stripping) rather than on developing their regions (Hoff and Stiglitz 2004). It would imply that the positive effect of local origin, in terms of improving the performance of governors, would be driven only by regions with patriotic elites. To deal with this problem, we pursue the following approach. First of all, we use the ‘patriotism’ variables presented above (in section 5.2) and also look at corruption (experienced and perceived) as proxies for rent-seeking and average amortization of assets in the region in 2010 as a proxy for asset-stripping. For each variable we compute the mean and create two dummies: the first one is equal to 1 for governors with local origin and a high value of patriotism / low values of corruption / low value of amortization (and zero otherwise); the second dummy is equal to 1 for governors with local origin and low value of patriotism / high value of corruption / high value of amortization (and zero otherwise). Then we run two triple interaction term regressions for each of the variables, using two dummies separately instead of the original local origin variable (and also replacing all interaction terms respectively). The results show us whether the effect of local origin is driven by regions with patriotic elites, or not. We find that the effect of local origin is present in both regions with high and low values of the patriotism variables; high-corruption and low-corruption regions; regions with high and low amortization of assets. Thus, our results are not driven by a particular type of elites. The only exception is the variable measuring the number of predecessors: it is only significant for regions with a high number of predecessor what, as mentioned, rather indicate powerless and less motivated elites.

Table 1: Correlation between local origin of governors and state capacity in Russian regions, OLS, dep. var.: local origin index Log bureaucracy Share of bureaucrats with university education Share of bureaucrats participating in professional training Number of agencies Number of municipalities Efficiency of public sector I (Aleskerov) Efficiency of public sector II (Aleskerov) Efficiency of public sector III (Aleskerov) Efficiency of public sector (Ministry of Regional Development) Improvement of public sector efficiency (Ministry of Regional Development) Experienced corruption Perceived corruption Democracy Freedom of the press

(1) -0.045 (0.194)

(2) -0.19 (0.211)

(3) -0.334 (0.272)

(4) -0.233 (0.226)

(5) -0.249 (0.236)

(6) -0.246 (0.290)

(7) 0.067 (0.235)

(8) -0.173 (0.224)

(9)

(10)

(11)

(12)

(13) 0.168 (0.524)

-0.011 (0.039)

-0.013 (0.039)

-0.017 (0.039)

-0.028 (0.040)

-0.01 (0.041)

0.02 (0.050)

-0.04 (0.041)

-0.032 (0.039)

-0.075 (0.079)

-15.209** (7.235)

-21.530*** (8.052) 0.011 (0.013)

-21.383** (8.101) 0.011 (0.012) 0.001 (0.001)

-23.281*** (8.042) 0.009 (0.012)

-21.952*** (8.001) 0.009 (0.013)

-22.989** (10.923) 0.013 (0.014)

-16.164* (8.504) 0.009 (0.012)

-22.320*** (7.936) 0.013 (0.013)

-9.675 (17.224) 0.011 (0.026) -0.001 (0.001)

0.017 (0.145)

0.108 (0.204)

-0.048 (0.150)

0.044 (0.190)

-0.229 (0.148)

-0.205 (0.189)

-0.104 (0.135)

-0.224 (0.217)

-0.007 (0.132)

0.179 (0.164) -0.729 (1.486) 3.655 (3.913) -0.079** (0.037)

-1.067 (0.991) 1.487 (2.076) -0.061** (0.027) -2.109 (1.666)

-0.003 (0.127)

Tax revenue to GDP Income tax revenue to GDP

0.029 (0.047) 0.05 (2.104) 0.000 (0.000) -0.026 (0.041)

Tax arrears to GDP Relative extractive capacity Crime rate Infant mortality

0.118 (0.192) -0.299*** (0.104) 0.063 (0.038) 0.344 (0.437) 0.000 (0.000) -0.000 (0.044)

Quality of forest governance: organization Quality of forest governance: law Quality of forest governance: economic efficiency Quality of forest governance: ecological performance

-0.015 (0.096) -1.547 (3.012) 0.000 (0.001) -0.045 (0.127) -0.016 (0.097) 0.072 (0.098)

-0.023 (0.098) 0.074 (0.100)

0.06 (0.117) 0.156 (0.171)

0.071 (0.084)

0.074 (0.085)

0.094 (0.171)

0.050 (0.101) -0.232 (0.325)

0.071 (0.137) -0.309 (0.381) 0.000 (0.000) -0.488 (1.154) 13.609 (9.900) 63 0.356 1.36

0.308 (0.564) 4.466 (5.796) 79 0.060 1.14

0.128 (0.581) 7.201 (6.081) 77 0.088 1.74

0.186 (0.585) 8.257 (6.251) 77 0.095 1.65

0.083 (0.591) 10.28 (6.483) 77 0.119 2.11**

0.198 (0.599) 7.499 (6.726) 77 0.097 1.40

-0.598 (0.759) 8.866 (6.478) 67 0.107 1.37

0.445 (0.591) 6.785 (5.925) 77 0.156 2.57**

-0.164 (0.533) 11.014* (5.632) 76 0.116 2.05*

0.222 (0.553) 2.145 (2.778) 79 0.007 0.15

-0.108 (0.555) 4.474 (2.803) 79 0.081 1.68

0.326 (0.519) 0.993 (2.791) 73 0.036 0.45

0.050 (0.102) -0.210 (0.325) 0.000 (0.000) 0.267 (0.530) 1.282 (2.829) 73 0.038 0.38

1.52 0.23

2.01 0.34

1.90 0.59

2.32** 0.71

1.63 0.35

1.37 0.80

3.08** 0.80

2.08* 0.50

0.17 0.15

2.01* 0.75

0.44 0.45

0.37 0.38

1.41 0.87

0.07

0.41

0.66

0.77

0.32

0.87

0.57

0.66

0.17

0.89

0.44

0.37

0.91

No independent forest agency Forest fire experience Log GDP per capita Constant Observations R-squared F-test (all covariates) F-test (all covariates except log GDP) F-test (all insignificant covariates) F-test (all insignificant covariates except log GDP)

Note: robust Huber-White standard errors in parentheses. *** 1% significance level, ** 5%, * 10%. Significant results marked bold. Using ordered logit does not change the results of regressions (except democracy becomes insignificant)

Table 2: Correlation between federal connections of the governor and state capacity in Russian regions, OLS, dep. var.: federal connections dummy Log bureaucracy Share of bureaucrats with university education Share of bureaucrats participating in professional training Number of agencies Number of municipalities Efficiency of public sector I (Aleskerov) Efficiency of public sector II (Aleskerov) Efficiency of public sector III (Aleskerov) Efficiency of public sector (Ministry of Regional Development) Improvement of public sector efficiency (Ministry of Regional Development) Experienced corruption Perceived corruption Democracy Freedom of the press

(1) -0.016 (0.069)

(2) -0.044 (0.074)

(3) -0.04 (0.105)

(4) -0.035 (0.078)

(5) -0.038 (0.080)

(6) 0.008 (0.085)

(7) -0.043 (0.096)

(8) -0.037 (0.078)

(9)

(10)

(11)

(12)

(13) 0.207 (0.219)

0.003 (0.017)

0.000 (0.017)

0.000 (0.017)

0.001 (0.017)

0.008 (0.018)

-0.006 (0.021)

0.000 (0.017)

0.003 (0.018)

-0.019 (0.035)

1.685 (3.120)

-1.378 (2.613) -0.002 (0.005)

-1.382 (2.620) -0.002 (0.005) 0.000 (0.000)

-1.149 (2.737) -0.002 (0.005)

-1.712 (2.633) -0.000 (0.005)

-0.450 (2.780) -0.002 (0.005)

-1.357 (2.885) -0.002 (0.005)

-1.240 (2.655) -0.002 (0.005)

7.103 (5.312) 0.000 (0.009) 0.000 (0.000)

0.049 (0.044)

0.043 (0.076)

-0.055 (0.063)

-0.043 (0.082)

0.049 (0.049)

0.006 (0.084) 0.006 (0.041)

0.102 (0.072)

-0.083* (0.047)

-0.106 (0.068) 0.297 (0.471) 0.241 (1.342) 0.013 (0.014)

0.124 (0.313) -0.527 (0.974) 0.000 (0.011) 0.278 (0.749)

-0.063 (0.049)

Tax revenue to GDP Income tax revenue to GDP

-0.011 (0.014) 1.106 (0.846) 0.000 (0.000) 0.005 (0.016)

Tax arrears to GDP Relative extractive capacity Crime rate Infant mortality

-0.032 (0.085) 0.028 (0.047) -0.009 (0.013) 0.067 (0.136) 0.000 (0.000) 0.005 (0.019)

Quality of forest governance: organization Quality of forest governance: law Quality of forest governance: economic efficiency Quality of forest governance: ecological performance

0.020 (0.033) 0.423 (1.353) 0.000 (0.000) -0.010 (0.043) 0.019 (0.034) -0.008 (0.034)

0.02 (0.034) -0.008 (0.035)

0.026 (0.049) 0.033 (0.053)

0.065** (0.027)

0.064** (0.028)

0.128** (0.061)

-0.013 (0.035) -0.108 (0.097)

-0.014 (0.044) 0.034 (0.164) 0.000 (0.000) 0.107 (0.441) -1.835 (4.829) 63 0.272 0.47

-0.361** (0.179) 1.81 (2.203) 79 0.051 1.73

-0.375** (0.182) 2.76 (2.165) 77 0.067 1.34

-0.377** (0.182) 2.728 (2.287) 77 0.067 1.11

-0.368** (0.183) 2.392 (2.317) 77 0.09 1.35

-0.284 (0.187) 1.686 (2.294) 77 0.116 1.24

-0.127 (0.221) 1.493 (2.334) 67 0.025 0.34

-0.374** (0.182) 2.758 (2.191) 77 0.067 1.11

-0.296* (0.168) 1.903 (2.253) 76 0.044 0.97

-0.205 (0.180) 1.343 (0.955) 79 0.085 1.37

-0.279 (0.205) 1.651 (1.093) 79 0.069 1.31

-0.323* (0.170) 1.781* (0.923) 73 0.138 2.05*

-0.013 (0.036) -0.110 (0.101) 0.000 (0.000) -0.315* (0.173) 1.746* (0.926) 73 0.139 1.46

0.19 0.19

0.20 0.20

0.16 0.16

0.95 0.95

0.89 0.81

0.14 0.34

0.19 1.11

0.15 1.42

0.70 1.37

0.62 1.31

1.78 0.85

1.48 0.70

0.48 0.44

0.19

0.20

0.16

0.95

0.19

0.14

0.19

0.16

0.70

0.62

0.37

0.29

0.45

No independent forest agency Forest fire experience Log GDP per capita Constant Observations R-squared F-test (all covariates) F-test (all covariates except log GDP) F-test (all insignificant covariates) F-test (all insignificant covariates except log GDP)

Note: see Table 1. Using logit, most major results are confirmed: in the specification (13) we also obtain multiple additional significant results, but the estimation suffers from perfect prediction problem and thus cannot be interpreted in a reasonable way.

Table 3: Correlation between local origin of the governors and the features of regional elites, OLS, dep.var.: local origin index (1) (2) (3) (4) (5) Distance from Moscow 0.030 0.027 0.031 0.020 0.02 (0.052) (0.053) (0.053) (0.053) (0.053) Oil and gas 0.000 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) Tenure of the predecessor -0.000 -0.000 -0.000 -0.000 -0.000 (0.000) (0.000) (0.000) (0.000) (0.000) Number of predecessors -0.294*** -0.294*** -0.314*** -0.285*** -0.286*** (0.095) (0.096) (0.094) (0.097) (0.096) Dummy republic 0.349 (0.281) Share of ethnic Russians -0.473 (0.617) Dummy Northern Caucasus -0.109 (0.486) Ethnic fractionalization 0.897 (0.562) Polarization 0.662 (0.521) Constant 3.851*** 4.291*** 3.961*** 3.673*** 3.651*** (0.371) (0.628) (0.363) (0.420) (0.438) Observations 79 79 79 79 79 R-squared 0.104 0.096 0.088 0.108 0.103 F-stat (all covariates) 2.55** 2.35** 2.31* 3.07** 2.72** F-stat (all insignificant covariates) 1.28 1.15 1.11 1.49 1.32 Note: see Table 1. Using ordered logit we can confirm almost all results, except in some specifications tenure of the predecessor becomes marginally significant and negative. Table 4: Correlation between federal connections of the governors and the features of regional elites, OLS, dep. var.: federal connections dummy (1) (2) (3) (4) (5) Distance from Moscow -0.018* -0.021** -0.017* -0.021** -0.019** (0.010) (0.010) (0.010) (0.010) (0.010) Oil and gas -0.000*** -0.000*** -0.000*** -0.000*** -0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) Tenure of the predecessor 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) Number of predecessors 0.135*** 0.141*** 0.140*** 0.136*** 0.134*** (0.039) (0.039) (0.038) (0.039) (0.039) Dummy republic 0.074 (0.117) Share of ethnic Russians -0.259 (0.230) Dummy Northern Caucasus 0.384* (0.196) Ethnic fractionalization 0.180 (0.263) Polarization 0.067 (0.221) Constant -0.303*** -0.091 -0.363*** -0.337** -0.313** (0.113) (0.229) (0.114) (0.130) (0.134) Observations 79 79 79 79 79 R-squared 0.151 0.166 0.207 0.151 0.146 F-stat (all covariates) 4.04*** 4.28*** 5.07*** 4.00*** 3.95*** Note: see Table 1. Using logit we can confirm almost all results, except ethnic fractionalization becomes significant and positive and share of ethnic Russians significant and negative.

6. Extreme bounds analysis A certain limitation of our robustness checks is that we are unable to estimate the regression including each and every control variable (which would be the best possible way to check for the omitted variable bias), because the number of variables compared to the number of observations in our regressions is simply too high, while many variables are also correlated. As a partial solution to this problem we use the extreme bounds analysis (EBA). Using this technique (in the modification suggested by Sala-i-Martin 1997), we identify the set of the most robust control variables (i.e., those which keep sign and significance in most regressions), run triple interaction term regression using these controls only, and plot the marginal effects. The obtained figure is, however, almost entirely identical to our main result. We start by using a set of 99 possible control variables, specifically: (1) territory; (2) mobility of the governor; (3) dummy governor studies not in the region of birth; (4) dummy governor studies in Moscow or St. Petersburg; (5) efficiency of regional government according to the Ministry of Regional Development; (6) governor studied at a top university (Web ranking); (7) governor studied at a top university (HSE ranking); (8) dynamics of efficiency of the regional government according to the Ministry of Regional Development; (9) – (11) three indices of bureaucratic quality from Aleskerov et al. (2006); (12) dummy for governors with rural origin; (13) freedom of the press; (14) democracy; (15) experienced corruption; (16) distance from Moscow; (17) oil and gas; (18) training of bureaucrats; (19) tenure of the governor’s predecessor; (20) number of predecessors; (21) polarization; (22) dummy governor of non-Russian ethnicity; (23) bureaucrats with university degree; (24) share of forest; (25) forest expenditures; (26) federal transfer for fire prevention 2010; (27) federal transfer for fire prevention in Q3 2010; (28) rainfall; (29) temperature; (30) GDP; (31) investments; (32) federal transfers; (33) road coverage; (34) crime rate; (25) dummy head of the region of the titular ethnicity of an ethnic republic; (36)-(38) profession of the governor; (39) business connections of the governor; (40) share of ethnic Russians; (41) ethnic fractionalization; (42) dummy republic; (43) dummy Northern Caucasus republic; (44) member of the United Russia; (45) held a regional office before; (46) dummy for governors who have never been elected; (47) number of electoral victories; (48) number of appointments; (49) wind velocity; (50) temperature deviations; (51) rainfall summer 2010; (52) tenure; (53) returned from holidays; (54) elections in October; (55) age; (56) state of emergency; (57) temperature summer 2010; (58)-(77) dummies for federal districts and measures of spread of various tree types; (78) dummy for governors born in the region of office; (79) average GDP growth rates in the region; (80) index of participation of local interest in forest governance; (81) – (84) proxies of governor’s experience in managing disasters / forest fires and large bureaucracies; (85) size of bureaucracy; (86) number of agencies; (87) number of municipalities; (88) tax revenue to GDP; (89) perceived corruption index; (90) tax arrears to GDP; (91) relative extractive capacity; (92) income tax to GDP; (93) infant mortality; (94) – (97) indices of quality of forest governance; (98) dummy for regions without an independent forest agency; (99) experience of regional bureaucracy in fighting forest fires. Then we use all possible combinations of these variables by four. It gives us 3,764,376 regressions, which we estimate one by one. Based on this analysis, we compute the CDF(0) statistic suggested by Sala-i-Martin (1997). We assume that the estimations are normally distributed and assign equal weight to the regressions. The following variables turn out to have CDF(0) exceeding 0.95: (1) studies at top-30 university; (2) freedom of the press; (3) spread of ash; (4) accumulated number of forest fires during governor’s tenure; (5) average annual spread of forest fires during governor’s tenure; (6) bureaucratic experience with

fighting forest fires; (7) number of appointments; (8) area; (9) distance from Moscow; (10) road coverage; (11) dummy Far East; (12) dummy North West; (13) dummy state of emergency; (14) spread of larch; (15) spread of beech; (16) infant mortality; (17) federal transfers for forest fire protection in 2010 and (18) federal subsidies for forest fire protection in Q3 2010. Since the variables (17) and (18) measure the same issue and variables (4)-(6) are highly correlated by design, we construct six combinations of these control variables. We then run six baseline regressions with a triple interaction term controlling for one combination and all other robust variables (we include population, federal connections, local origin and their interactions). All regressions confirm our results.

7. Vice-governors In this section we look at the biographies of vice-governors, the second-highest bureaucrats in the regions. Vice-governors in Russia are appointed by governors and replaced more frequently than governors. During the period of our investigation, only 6 regions had vicegovernors with longer tenure than incumbent governors, and in three of these regions the governors were appointed less than a year ago (probably, they simply had no time to reappoint their second-in-command). Moreover, unlike lower-level bureaucrats, who are almost without exception of local origin, vice-governors are also occasionally outsiders. It is reasonable to hypothesize that outsider governors are more likely to appoint outsider vicegovernors. If that is the case, one could hypothetically assume that our results are not driven by the knowledge of governors, but by the lack of local knowledge of vice-governors. For this purpose we have compiled a dummy equal to 1 for regions with vice-governors of local origin and zero otherwise (unfortunately, more detailed data are unavailable for this group of bureaucrats). 7.1 Method In this robustness test we investigate the local origin of regional vice-governor. We use the publically available biographies of vice governors in order to collect individual data and determine their local origin respectively. For this purpose we have set up a dummy variable, which is one, whenever there is a vice-governor with local origin in office during the wildfire months (July-September 2010) and zero otherwise. Similar to the binary local origin variable for regional governors, the vice-governors with a local origin value of zero have never lived in their region of office (at the year of appointment), or moved briefly before their appointment to the respective region of office (we accept 1-2 years as a threshold). More detailed information is, unfortunately frequently unavailable. 7.2 Assumptions The data on vice-governors is somehow different from the governor data. First, we had to look carefully in order to identify the ‘second most powerful’ bureaucrat in the region. Many Russian regions have a unique governance structure. For example, in some regions the second highest bureaucrats are literally the vice-governors. Other regions differentiate between the first vice-governor and other vice-governors. In other regions the position of the vicegovernor is absent and the second highest bureaucrat is the head of the regional government (in some cases, again, the governor is automatically the head of government, in other regions this position is occupied by a different person). Multiple other alterations can be observed. Another problem we faced was that some regions have more than one vice-governor (because they have two ‘first vice-governors’ or no hierarchy among vice-governors at all). For these regions we use the following rule. If at least half of the vice-governors (with equal rank, so without, for example, differentiation between ‘first’ and ‘regular’ vice-governor) or more are of local origin, we assign a local origin score of one to the region. 7.3 Results We find that in regions run by outsider governors outsider vice-governors are significantly more frequent25 (in 50% of the cases, outsider governor has an outsider vice-governor; while 25

To make the data on governors and vice-governors comparable, for governors we also use the binary local origin index.

only 10% of local governors had an outsider vice-governor). However, if we look at regions run by outsider governors, and compare the average forest fire spread for regions with and without local vice-governors, no statistical difference could have been established (the average spread of a forest fire is almost identical). The same is true for regions run by insider governors – here the spread of forest fires does not depend on whether vice-governor is of local origin or not. Thus, it looks like it is indeed the local origin of the governors themselves which is driving our results.26

26

We also replicated the baseline regressions controlling for the local origin of vice-governors. The results remained robust, but should be treated with caution: as mentioned, vice-governors are typically appointed by governors, and thus by controlling for their local origin we may encounter the post-treatment bias problem.

8. Case study: Ryazan region 8.1 Ryazan region during the wildfires The wildfires of 2010 left Ryazan region in a devastated condition. The Russian statistical agency reported 236 forest fires covering a total of 137,183 hectares of forest area, burning down over 269 houses and killing 8 people.27 During the wildfires approximately 4,000 people with 350 units of technical equipment were at work, including units from 8 other Russian regions, the army, the federal ministry of emergency situation, aviation and personnel from Azerbaijan, Turkey, Poland and Belarus and uncounted volunteers.28 In addition, Ryazan received about 100 thousand tons of humanitarian aid.29 The emergency situation status declared by the federal government lasted 3 weeks, longer than in any other Russian region. 8.2 A governor without local origin In 2008 Oleg Kovalev was appointed governor of Ryazan region. He was born in Krasnodar (Southern Federal District), studied in Saratov (Volga Federal District), after which he worked in two federal ministries in Moscow City. After 1991 he worked in the administration of a small district of Moscow region and in 1999 was elected to the Russian parliament (Duma). After several re-elections he was appointed governor of Ryazan region in 2008.30 After his appointment, Kovalev partly staffed his regional government with loyal followers and other outsiders. In May 2008 he appointed Valery Ionov to his vice-governor and in December 2008 promoted him to the rank of first vice-governor of Ryazan region (Ionov has a background in engineering and finance). Similar to the governor Kovalev, vice-governor Ionov never lived in Ryazan before his appointment to the regional administration. He was working in a district administration of Moscow region (where he probably met Kovalev), after which he worked in various financial companies in Moscow City and region.31 As first vicegovernor Ionov was responsible for public property, land ownership, construction, agriculture and forestry. At the time Ionov’s appointment, the federal government decided to decentralized forest management to the regional level and in Ryazan region he happened to be the responsible bureaucrat to implement the decentralization of forest fire safety (among others). As a consequence, in 2008 the regional public agency ‘GU RO Pozhlez’ was founded and assigned the task of wildfire prevention and extinction. Pozhlez was financed by the regional budget and endowed with four fire fighting vehicles, three tractors, 16 fire-chemical stations and 142 personnel (as of 2010 before the wildfires),32 and thus seems to be extraordinary small compared with 1013 thousand hectares of regional forest area.33 The agency was headed by Victor Vaskin, originally from Ryazan region, but inexperienced in fire prevention and extinction as he worked as a manager in the regional commercial forestry sector (he was most likely selected and appointed by Ionov). Thus even though governor Kovalev affirms in 2009 that “forest fires are the problem number one for me”34 and in 2010 that “during the last two years as governor, I have spared funds 27

Rosstat report “Forest fires”; RosBusinessConsulting (19.08.2010). Moscow Carnegie Center, Ryazan region: July-August (2010) by Vadim Abramov. 29 Ibid. 30 Because of his long stay in Moscow City and region, which are, similar to Ryazan region, part of the Central federal district, we have assigned governor Kovalev received the score 2 of the local origin index. 31 Specifically, Ionov worked in the Podolsk district administration which is close to the Kashirsky district, Kovalev`s previous workplace. Both districts are in Moscow region. 32 Ryazan Vedomosti, 09.11.2010. 33 Rosstat 34 Mediaryazan, 10.06.2009 28

from our scare budget in order to strengthen Pozhlez, our central forestry administration”35 he does not deal with forest management directly. He has to rely on appointed bureaucrats; as we will show in what follows, in case of Ryazan this mechanism did not function. 8.3 Pre-crisis evidence In May 2009 the prosecutor of Ryazan investigated whether the district administrations and the regional branch of the ministry of emergency situation fulfill their legal obligations in the area of fire safety and are able to extinguish potential forest fires. The investigation has shown that local bureaucrats insufficiently implement fire prevention and surveillance measures resulting in administrative and disciplinary sanctions for more than hundred local bureaucrats (e.g. missing fire warning signs and gates preventing access to forests; recreation resorts unequipped with water tanks and hoses; etc.).36 Moreover, the prosecutor requested the executives of Pozhlez to purchase the necessary equipment for firefighting and to repair the broken machinery.37 Thus, already before crisis happened some information on potentially low performance of regional administration in dealing with forest fires was available. Still, little was done before the fire crisis actually happened. 8.4 A timeline of events: wildfires in Ryazan region At the end of June 2010, during a period of enormous heat and drought, governor Kovalev ordered the district heads and executives of various state agencies (regional branch of the ministry of emergency situation, police, forest agency), in a meeting of the executive committee of the regional government to arrange appropriate measures wildfire prevention and prepare personnel and equipment for possible operation to extinguish emerging wildfires.38 Specifically, he ordered to organize additional forest patrols (especially at vacation resorts and at lakes), restrict the access to forest, abstain from carrying out public events close to forests and prepare technical equipment necessary for fire extinction in the case of emergency. However, none of the directives were implemented. When the wildfires started, the bureaucrats showed little effort in improving their performance. There are numerous blog entries of eye witnesses during the wildfires that emphasis this fact.39 For example, people complained that some local executives were not accessible because they were on summer holidays and their temporary substitutes were unable to cope with the situation (even the chairman of the regional parliament was abroad for holidays); local bureaucrats did not take incoming complaints/requests/clues seriously. Even after directives of the governor were published locals witnessed crowed forest areas.40 Thus, while the governor seems to make ‘right’ decisions (and to make them ‘on time’), he fails at enforcing them and overseeing their implementation by the regional bureaucracy at the lower level. In addition to being misinformed about the behavior of bureaucrats, Kovalev seems to be misinformed about the extent of forest fires themselves. By mid of July 2010 the first serious forest fires emerged in Ryazan, but only by the end of July when the wildfires threatened the first villages and urban areas the regional government reacted. On the 28th of July the governor reaffirmed his directive to prohibit access to forests and recreational activities close to forests. On the 30th of July Kovalev finally declared the regional state of emergency 35

Mediaryazan, 25.08.2010 InformRyazan, 26.05.2009 37 Ibid. 38 Rzn.info, 29.06.2010 39 Forest fires in Ryazan region: http://forum.rzn.info 40 For example: http://lun198.livejournal.com/2692.html 36

creating higher sanctions for those who risk forest fire safety (e.g. lighting bonfires in forests). At this point in time numerous houses were burned and enormous forest areas have been destroyed. One possible explanation for this delay could be an incorrect assessment of the situation in the region. To provide a comparative case, in Yaroslavl region which is also in Central Russia and close to Ryazan region the governor (who has a high local origin score) declared the regional state of emergency already on the 24th of July. On the 30th of July governor Kovalev convened an emergency committee headed by firstvice governor Valery Ionov and vice-governor Tatiana Panfilova (a local physician who entered politics in 2001 and was appointed vice-governor in 2005) to centrally organize the liquidation of forest fires. It is appealing that no local bureaucrats with actual forest fires experience were selected; it may again reflect poor knowledge of which human resources regional bureaucracy actually had at its disposal (particularly problematic due to the need to quickly react to forest fires). At the end of July Kovalev changed his strategy. Instead of issuing seemingly ineffective directives for local bureaucrats he focused his effort to source support from outside. He requested help from Sergey Shoygu (minister of emergency situation), Vladimir Putin (prime minister) and other top-tier bureaucrats (e.g. chairman of the Federation Council). On the 2nd of August President Medvedev declared the federal state of emergency on Ryazan region which initiated large-scale external support units (from other Russian regions, abroad and volunteers) for the region which eventually were able to extinguish most of the wildfires by the end of August. Only later, the negligence and disorder of Ryazan`s local bureaucracy appeared evident. Thus in September a number of bureaucrats were sacked after it became known that in one district 15 tons of humanitarian aid for victims of the forest fires was literally thrown away. Among the fired bureaucrats were the minister and first vice-minister of the regional ministry for social protection of the population. Moreover, the regional prosecutor has opened a criminal investigation with the charge of negligence against a number of local bureaucrats in another district of Ryazan region, as a consequence of the absence of effective forest fire safety measures which lead to the destruction of 52 houses leaving 125 people without shelter and killing one person. Summing up, the disastrous outcomes of forest fires in Ryazan were driven by two factors. First, the governor was unable to properly monitor regional bureaucrats (thus most of his decisions were ignored at least until the peak of disaster) and systematically selected officials without appropriate background to deal with forest fires. Apparently he lacked local knowledge of high-level bureaucrats and regional elites. Second, the governor also seems to lack appropriate information about the extent of fires due to miscommunication of local bureaucracy. Although Kovalev managed to procure federal support, it was obtained too late when the damage from the disaster was already excessive.

.

9. Governors` tenure and regional residence 9.1 Method While determining the tenure of the governors, we encounter the following problem: the exact moment of power (authority) transfer cannot be determined consistently for all governors. First of all, there is a variation of how governors are selected over time. Many of the longlasting governors which came into power before 1997 have been appointed by President Yeltsin. In the subsequent period, until 2004, governors were elected by the regional population. After 2004 governors were appointed again. Moreover there is a small intraregional variation in the appointment routine of governors. For example, after 2004 the official appointment process has three administrative stages. First, the president proposes a candidate to the regional parliament (in case the incumbent governor voluntarily retreats the president will appoint an interim governor until he offers a suitable candidate to the regional parliament, while in most cases the interim governor turns out to be the president’s man). Second, the regional parliament approves the candidate as governor by election. Finally, there is an official inauguration ceremony in which the new governor formally assumes the office (the ceremony can take place one or two months after the election victory or acceptance by the regional parliament). The sequencing of the three stages differs. Sometimes all three stages may happen within one day, while in other cases the regional administration abstains from an official inauguration ceremony at all. In order to deal with both sources of variation and to stay consistent for all governor we consider the start of a governor’s tenure at the date of the appointment decree (for governors who have been appointed by Yeltsin), the date of election victory (for elected governors until 2004), as well as the date of the official acceptance of the president’s candidate by the regional parliament (for governors appointed by Putin or Medvedev). An additional difficulty results from the fact that in many regions the election and inauguration of the regional governor takes place at the end of the year (and sometimes the events fall in two successive years). In other regions governors are replaced in summer or in spring. In order to increase precision we look at days in office, starting from date of appointment (as defined above) until July 21st, 2010. We pick this specific date for two reasons. First, it is widely accepted that the Russian wildfires in 2010 started in late July. Second, by adhering to our previous “appointment definition” we have to consider the specific case of Karelia region (Northwestern federal district). On the 30th of June 2010 the incumbent governor of Karelia resigned and Andrey Nelidov was appointed as “temporary acting” governor until a new candidate is found. Eventually, at the 17th July president Medvedev proposed Nelidov as governor candidate to the regional parliament. On the 21st July Nelidov was elected by the parliament and directly inaugurated. Therefore, in order to stick to our tenure calculation logic we set the 21st of July as general end date for the governors` tenure calculation. 9.2 Results If we introduce tenure duration in the set of covariates, we establish unambiguously that it does not affect the performance of the governors. We do not find any effect of this variable individually and interacted with urban population even after controlling for age and local origin (two-way interaction term). In a triple interaction after substituting tenure for local origin, we observe a positive effect of federal connections on forest fire expansion conditional

on tenure periods between 3-7 years in regions with small urban population (see Figure 8).41 However, the fact that short (0-3 years) and long tenure periods (7-18 years) turn out to be insignificant, as well as the fact that the gubernatorial re-appointments generally occur after 4 (or 5) years indicates that the positive effect of federal connections is unlikely to be driven by knowledge accumulation during the tenure period, but instead by governors with upcoming (or recent) re-appointments who enjoy increased attention by the federal administration.42 Thus, we can argue that tenure does not matter for the performance of regional governors. On the other hand, the previous effects of federal connections and local origin are robust to the inclusion of the tenure variable in the set of controls (as well as the corresponding interaction term with urban population and its squared term).43 It indicates that the accumulation of local knowledge before the appointment as governor is more important than the experience as governor in the region. There are two possible interpretations for this result. First, since the average age of governors in our sample is above 50 years, and the average tenure is slightly above 5 years, the time the governors spent in office is simply too short as opposed to their pre-appointment life experience for most of them. Consequently, when we substitute a continuous local origin variable capturing the number of pre-inauguration years of regional residency for the local origin index we find an increasing and positive effect of federal connection starting from 10 years pre-inauguration residence up to 50 years of residence (see Figure 8).44 Second, as mentioned, if the main type of local knowledge affecting the performance of the governor is the ‘knowledge of regional elites’, there are reasons to expect that the accumulation of this type of knowledge over the duration of tenure is more difficult – and this is precisely what we observe in our regressions.

dio=500

1.0e-04

-4

dio=6000

Density 2.0e-04

dio=4500

3.0e-04

dio=3000

* * * * * * * * * * * * * * * * * *

0

-8

-6

Margina effect

-2

0

dio=1500

4.0e-04

Figure 8: Tenure and regional residence

10

12

14

16

log urban population

Marginal effect of federal connections on forest fire management effectiveness conditional on the size of urban population for different levels of tenure (dio=days in office), triple interaction term

41

0

2000

4000 daysinoffice

6000

Distribution of tenure

The results are robust if we control for local origin and age. On the other hand, tenure has no significant effect conditional on federal connections for various levels of urban population. 42 We also test for then effects of local origin conditional on tenure and urban population (two-way and triple interaction). However, local origin has no effect conditional on tenure duration (specifically, we test for office periods of 1, 4, 8, 12 and 18 years). 43 The same holds when controlling for governors` age (individually, as interaction with urban population and as squared term). 44 The results hold irrespective of whether we control for tenure and age of the governor.

.04

2

YR=50 YR=40

Density .02 .01

-2

YR=10

* ** * * * * ** * * ** *** ** * ** ** * * * * * * * ** * * * * * * * * * * * * * * * * * * * * *

0

-6

-4

Margina effect

0

YR=20

.03

YR=30

10

12

14

16

Log urban population

Marginal effect of federal connection on forest fire management effectiveness conditional on the size of urban population for different levels of regional preinauguration residence, triple interaction term

0

10 20 30 40 50 Years of regional residence (as of year of inaguguration)

Distribution of years of regional residence (excluding gubernatorial tenure)

Notes: The red stars above the lines indicate whether the marginal effects of tenure (or regional pre-inauguration residence respectively) are significant at 5% for the respective range of log urban population (or in other word whether the confidence interval around the lines are either completely above or below the zero axis).

10. Governors’ talents and experience 10.1 Education and mobility It is possible that local origin and federal connections reflect a different characteristic of the governors other than local knowledge and the presence of links to the federal government. Specifically, they may reflect governors’ intelligence, organizational talents, ‘soft skills’ and motivation. For instance, ‘smarter’ governors could be aware of the advantages of federal connections in a highly centralized federation and therefore made their career choice accordingly (some sort of self-selection of more qualified individuals into public service). Then it is not clear whether better performance of the governors with a background in the federal bureaucracy is due to the presence of federal connections or determined by their intelligence, driving both federal connections and behavior during forest fires. The same argument could be used with respect to the local origin. However, in this case the interpretation seems to be less likely due to the institutional specifics of the Russian Federation. Our results indicate a higher quality of policies implemented by local governors compared to outsiders. Thus, one had to conclude that more ‘intelligent’ governors (‘intelligence’ here stands for all other ‘soft skills’ which could matter for a public official for simplicity) are more likely to stay in their region, while less ‘intelligent’ work elsewhere. This is, however, inconsistent with the patterns of mobility in Russia. Russia is a highly hierarchical country not only in terms of its political system, but also in terms of the educational system and the economic structure. For instance, the best universities are located in the large metropolitan centers, with Moscow being the absolute leader in almost all disciplines (followed by St. Petersburg and Novosibirsk). Thus, for most regions of Russia it is the case that the more talented high school graduates enroll in the programs outside their regions, while the less talented stay in the region. Similar logic applies to the further professional life: more successful specialists migrate to Moscow, less successful stay in their region. After the graduation, the best students rarely attempt to return to their home regions, aspiring a career in the capital cities. Horizontal migration between centers of equal importance is less often. Thus, for almost all parts of Russia (with the exception of Moscow and St. Petersburg, which are excluded from our sample anyway) it is safe to say that the more ‘intelligent’ are usually more likely to spend a substantial part of their career outside of their region. This contradicts our findings, which imply better performance of governors with ‘local origin’ making the interpretation that local governors are more ‘intelligent’ less likely. Still, the self-selection hypothesis of more ‘intelligent’ governors into particular career paths requires detailed analysis. Unfortunately, it is impossible to measure the governors’ intelligence directly (e.g. school grades, IQ test etc.) due to data availability, so we have to devise indirect measures of their soft skills. First, a possible indicator for the intelligence of a politician is the quality of his higher education and the reputation of the respective university. The governors in the sample have diverse educational backgrounds. While some graduated from the most distinguished Russian universities, others merely received some sort of vocational training. Second, another indicator for the personal characteristics of a governor contributing to his ability to effectively manage the affairs of his region might be his mobility. It is reasonable to assume that mobile individuals are more entrepreneurial and ambitious than those staying in a particular location for their entire life. For this purpose we look at the number of regions a governor worked in before the inauguration. Although the region of work and residence do not necessarily have to coincide, taking into consideration the size of many regions it is most likely. For both sets of variables we implement two types of tests: (1) a simple mean comparison (or, for ordered variable, correlation), showing whether governors

with federal connections or with local origin are more likely to have received better education or have been more mobile, and (2) re-estimate our main regressions controlling for the additional variables. The conclusions are unequivocal. We do not find any evidence of better education of governors with federal connection as opposed to other governors regardless of the proxy we use (see Table 5). We also find no difference in terms of local origin between governors, who have studied at top universities and other governors (see Table 6). For other measures of educational quality we find that the quality of education of governors with local origin is actually lower than of governors without local origin. Moreover, local origin and mobility are significantly and negatively correlated (correlation coefficient of minus 0.596). Governors with federal connections have higher mobility, probably, as an outcome of requirements of their career paths (see Table 7). Thus, we find little evidence that governors with federal connections or with local origin have higher quality of their human capital (except the evidence of higher mobility of federally connected governors). Furthermore, there is no evidence that governors with higher quality of education or more mobile governors performed better in terms of controlling the forest fires spread (see Table 8). Moreover, re-estimating our regressions controlling for education or mobility of governors does not change our results. We also use an alternative approach to check for the possible self-selection into federal bureaucracy. The fact that the federal bureaucracy offers attractive career opportunities became clear under Putin. However, the situation was entirely different under the previous president Boris Yeltsin, when business careers or jobs at regional administrations were more likely to be attractive due to extreme weakness of federal bureaucracy. The ascension of Putin to power was extremely fast and unexpected (it suffices to say that after his appointment as prime minister in 1999 he had a one-digit popularity rating significantly below another possible candidate for the presidency, Evgeny Primakov). Therefore, it is almost impossible to expect an individual to anticipate this shift in the 1990s and adjust her career path accordingly. Therefore, we checked for the work experience of governors with federal connections in the 1990s. If there was a self-selection of more talented and clever individuals into the federal service going on, one should expect all of them to have worked outside the federal administration in the 1990s, and join it in the 2000s, when the career options became evident. On the other hand, if a substantial fraction of these individuals worked for the federal government already in the 1990s, the self-selection becomes less likely. In our sample the career paths of the governors differed substantially: but we still find that 50% of the governors with federal connections entered the federal public service before 2000 (and even before 1999, when Putin became prime minister). 25% have worked for the regional governments and served in regional parliaments, 19% in private sector, and the rest in the military. Hence, there is very little evidence of self-selection going on, and the interpretation of the federal connections dummy used in this paper seems to be confirmed. Table 5: Federal connections, governors with different quality of education Variable 1 0 Ranking Web 0.273 0.182 Ranking HSE

0.100

0.209

Studies in Moscow or St. Petersburg

0.240

0.173

Studies not in the region of birth

0.256

0.118

Δ -0.091 (-0.698) 0.109 (0.804) -0.067 (-0.687) -0.138 (-1.523)

Note: Each entry represents the average value of federal connections dummy for governors, for whom the respective dummy for the quality of education is either 1 or 0. Each row corresponds to a particular education indicator. t-statistics in brackets. *** significant at 1% level, ** 5%, * 10% Table 6: Local origin, governors with different quality of education Variable 1 Ranking Web 2.909

0 3.030

Δ 0.121 (0.328) Ranking HSE 3.200 2.985 -0.215 (-0.559) Studies in Moscow or St. Petersburg 2.480 3.269 0.789*** (3.020) Studies not in the region of birth 2.721 3.382 0.661*** (2.651) Note: Each entry represents the average value of local origin score (1-4) for governors, for whom the respective dummy for the quality of education is either 1 or 0. Each row corresponds to a particular education indicator. tstatistics in brackets. *** significant at 1% level, ** 5%, * 10% Table 7: Federal connection and forest monitoring, difference of means in the level of mobility between governors with and without federal connections Federal connections 1 0 Δ Mobility (including Moscow and St. Petersburg) 2.750 2.095 -0.655** (-2.412) Mobility (excluding Moscow and St. Petersburg) 2.666 2.113 -0.553** (-0.200) Note: Each entry represents the average value of mobility (number of regions governor worked for) for governors, for whom federal connections are either 0 or 1. The first row computes the mobility including Moscow and St. Petersburg; the second excludes these regions. t-statistics in brackets. *** significant at 1% level, ** 5%, * 10% Table 8: Forest fire spread, governors with different quality of education Value of the education variable 1 0 Ranking Web 2.201 2.861

Δ 0.660 (1.159) Ranking HSE 2.400 2.815 0.375 (0.630 Studies in Moscow or St. Petersburg 2.588 2.852 0.264 (0.618) Studies not in the region of birth 2.962 2.519 -0.443 (-1.106) Note: Each entry represents the average spread of a forest fire for regions ruled by governors, for whom the respective dummy for the quality of education is either 1 or 0. Each row corresponds to a particular education indicator. t-statistics in brackets. *** significant at 1% level, ** 5%, * 10%

10.1.1 Method: education We look at the educational institution in which the governor has studied (and graduated) subsequent to his school education (which typically falls in the age of 20-25) and try to identify whether the respective university is among the top 30 Russian higher education institutions in terms of prestige in the country. We ignore any professional education and postgraduate degrees (e.g. PhDs) since we cannot be certain that the governors achieved the degree by their own accomplishment and not through informal support of staff or even bribes. In order to measure educational quality, four proxies are used. Two are associated with geography of Russian university education, where the best schools are located in Moscow, St. Petersburg and a handful of other big cities. Two are based on two university rankings, from which we derive a dummy for governors, who have studied at the best universities in Russia. In our sample we have no governor who has received their education in a Western country. One governor in our sample graduated from a Ukrainian institution. For him the dummies for

top universities are set to be equal to 0. Although this decision has not been made according to the ranking, it fits the above described logic, as the respective governor studied in a relatively unknown institution for vocational training. First, we base our analysis on the data of the Ranking Web of World Universities which includes 12,000 universities worldwide (450 Russian higher education institutions) and is based on an internet link analysis (for more information on the methodology see http://www.webometrics.info/). The ranking is compiled by a research group of the CCHS (part of CSIS, the largest public research group in Spain) and is updated every six months. We use the ranking of July 2011. The ranking has a major advantage in comparison to other international university rankings: if one looks at other international rankings (e.g. QS, or THE university rankings) one will hardly find any Russian university due to the relatively small sample size of the rankings (most of them include only the Lomonosov University in Moscow, but not a single governor has studied at this university, according to our data). The Ranking Web looks primarily on the Internet visibility of the universities, including not only their own “activity” in this respect, but also the overall attention to the university in the Internet. Clearly, it is not an objective measure of the university’s quality. But it does reflect the prestige of the university (in fact, even if the most prestigious universities do not care about their Internet appearance, they will still be actively discussed in the Internet due to their status). Prestige is, however, what we want to capture. As a result we use a dummy variable which is one if the governor has graduated from a one of the 30 universities and 0 if otherwise. We do not use the individual ranks of particular schools in ranking since, as mentioned, ranking is a rather crude measure, and it is necessary not to over-estimate the small differences within its scale. Second, we also use another university ranking compiled by a Russian institution – the Higher School of Economics (one of the leading Russian universities located in Moscow). The ranking was published in 2010 and is based on the score of the students admitted to the universities in their entrance exam (which in Russia also serves as the final exam for high schools). We use exactly the same procedure (creating a dummy for 30 top universities). The caveats mentioned below apply to this ranking as well, yet by using a different (also possibly imperfect) ranking we still can partly evaluate the robustness of our results. Since most of the governors received their education during the Soviet period, the names of the universities changed between their graduation and the period the ranking was prepared. We have tried to adjust for that. We also took the possible mergers of universities into account. We should acknowledge that the reputation of universities partly changed significantly between the Soviet and the current period, and some governors received their education in the USSR (for example, economics or legal studies gained at importance). This is a caveat one has to accept, as there are no rankings of the Soviet period available. A further problem is that some received their place at a prestigious university not because of personal talents or qualifications of any sort but rather because of connections of the parents. This type of informal connections plays a large role in the Russian educational system. Yet another problem is that prestige does not necessarily reflect the quality of training and the skills required to pass the program – while it is certainly the case for sciences and mathematics, in social sciences traditional Russian universities are often not very challenging. On the other hand, even if the program as such is not challenging, but the reputation of the school is high, it can still create a stimulating competitive atmosphere. Furthermore, prestige matters in terms of possible informal network formation, which may support the graduates in their future career and serve as yet another hidden factor influencing both the decision to join the federal public service and the success in managing the region.

10.1.2 Method: mobility While counting the regions of employment we use the following assumptions. First, we count the first region of employment regardless whether the governor has graduated in that region. Second, we count the region of gubernatorial appointment if the governor has not worked in that region before. The consideration behind the first two rules is that normally the locality of the first employment cannot not easily choose (especially not in the Soviet Union), while later decisions, especially on career development and location of employment is more deliberately chosen. Third, if a governor had various, however discontinuous work engagement in one region, we only consider the regions once. Fourth, an employment abroad is counted as ‘additional region’. There is one governor who worked in several regions (oblast) of Ukraine. We count the positions in different Ukrainian regions separately (Ukraine is subdivided into 25 regions). Fifth, governors who had a seat in the national parliament are counted as work position in Moscow. 10.2 Experience in civil service and disaster management Finally, we also have to deal with a further explanation of our findings. It is possible that the local origin or federal connections of the governors are correlated not with their generic intelligence or soft skills, but rather that they were associated with experience in governing functions and high-level management in bureaucracy, on the one hand, and with experience in disaster management, on the other. More experienced bureaucrats may find it easier to organize the work of public officials even without detailed knowledge of the regional elites; training in disaster management may be helpful for dealing with forest fires. In order to capture the experience of regional governors in managing a large number of bureaucrats we screen their publically available biographies and determine their first appointment in the public service which requires the coordination of a large crowd of civil servants. Such positions include vice-minister/agency (on the regional and federal level), vicegovernor, head of regional/federal government, vice-head of the administration of the president and the respective superordinate positions. We ignore positions on the district level and members of parliament due to the comparatively small number of subordinate bureaucrats. As a result we calculate the total number of years the governors are experienced in organizing a large number of bureaucrats until the forest fires in 2010 (experience in civil service management). Moreover, we create a dummy for three governors who should be accustomed with crisis situations: including one governor who worked in the Ministry of Emergency Situations and two governors who previously served in the army and have war experience (experience in disaster management). Basically, the list includes three governors, of which one has worked in the Russian Ministry of Emergency Situations and two have military combat and leadership experience in various wars, what may provide experience relevant for disaster management. None of the governors in our list is a trained disaster management professional. Finally, we use data provided by the federal forestry agency (through Rosstat) and calculate the total sum of yearly forest area covered by fire over the governors` tenure (and the yearly average) to test the governors` exposure to wildfires (forest fires during governors tenure and average forest fire during governors tenure). These proxies are likely to reflect the experience in forest fire management governor could have accumulated during his career.

In the first step, we regress the local origin and the federal connections on these variables. Local origin is indeed significantly and positively correlated with the experience in managing and coordinating large number of civil servants. This effect, however, is driven only by the fact that local governors, as mentioned, have generally longer experience in civil service. If we control for the tenure of the governor, the effect disappears. At the same side local origin is negatively correlated with experience in disaster management (since it is in Russia a federal function), again suggesting that the main benefit of governors with local origin is their knowledge of local elites and bureaucracies and not specific disaster management skills. Federal connections are weakly correlated with disaster management skills, but the result is driven only by Bashkortostan (once we exclude the region, the effect becomes insignificant). Controlling for any of these variables does not change our results. Table 9: Correlation between governor’s characteristics and governor’s experience in civil service management and disaster management, OLS (1) (2) (3) (4) Dep. var. Local origin Local origin Federal Federal connections connections Experience in civil service management 0.000 0.000 0.060*** 0.062*** (0.007) (0.006) (0.020) (0.020) Experience in disaster management -1.470*** -1.472*** 0.481* 0.491* (0.460) (0.458) (0.282) (0.284) Forest fires during governors tenure 0.000 0.000 (0.000) (0.000) Average forest fires during governors tenure 0.000 0.000 (0.000) (0.000) Constant 2.519*** 2.499*** 0.190** 0.171** (0.245) (0.254) (0.073) (0.071) Observations 79 79 79 79 R-squared 0.159 0.159 0.053 0.060 Note: see Table 1. Using logit and ordered logit yields identical results

11. Forest fire data in Russia The main specification of this paper was based on yearly forest fire data compiled by the Rosstat, the official Russian Statistical Agency. The data is based on report “5-LX: Records on forest fires” and is collected according to the following standardized procedure: Each legal entity in Russia that is required by law to implement measures against forest fires (hence on whose land forests are located), has to fill out report 5-LX and send it to its respective regional sub-unit of Rosstat. The report includes forest fire statistics until the 1st of November for the pending year and has to be submitted to the regional statistics office by the 10th of November. The report includes wildfires that happened on the land of the Russian forest fund as well as of land of other categories which include urban land (city forests), military land (military land), agricultural land etc. Moreover, the report considers forest fires caused by all reasons including unknown causes (such as agricultural burning, human causes, logging companies etc.). The Russian forest fund includes the natural reservoir of forests on the territory of the Russian Federation with is in federal ownership excluding forests in direct ownership of regions and municipalities, forests on military territory and urban settlements (e.g. forests within large territories of large cities), tree and shrub vegetation on agricultural and transport areas, and trees on the territory of the water fund (e.g. area of the water fund is the territory on which water objects are located including “water conversation zones and hydraulic facilities etc.). In mid-2012 the Federal Forestry Agency released a different dataset, which, among others, also include forest fire statistics and dates back to 2009. Despite the fact that the two datasets are compiled by different federal agencies using different standardized reports (Federal Forestry Agency uses report “6-OIP: Record on forest fires for different land categories of the forest fund, forest use, and fire types”), their major methodological difference is that the Rosstat dataset contains forest fires on the territory of the Russian forest fund and forest fires on “other territories”, while the Forestry Agency dataset solely comprises forest fires on territories of the forest fund. However, the deviations in the Federal Forestry Agency dataset should not be substantial since 99.6 percent of the Russian forest is located on territories of the forest fund (as of 1st of January 2010, 1183.7 million hectares land was covered by forests of which 1143.6 million hectares belong to the forest fund). Thus the dataset not only show a high correlation (e.g. correlation coefficient of 0.94 for the log dependent variables) but also in absolute term (e.g. number of forest fires in 2010 for Rosstat 34738 fires and for the Forestry agency 33345 fires). Furthermore the Forestry Agency distinguishes between forest area and non-forest area of the forest fund and reports separate indicators for these components. The former includes forest area covered by forest vegetation (protected, commercial and reserve forests) and forest area not covered by forest vegetation (but covered with bog, water, scree, roads, glaciers etc.). There are two characteristics which make the Federal forestry agency dataset particularly interesting for this paper. First, it includes information on the “non-forest area” of the forest fund which was covered by wildfires. Non-forest area is land which is unsuitable for forest vegetation (because it is covered by bogs, water, scree, roads, glaciers etc.). Second, the data is published on a quarterly basis, which makes more specific identification of timing of forest fires possible – this is of crucial importance for our analysis. 12.1 Quarterly distribution of forest fires If we scrutinize the distribution of forest fires on a quarterly basis (on the territory of the forest fund), we see that there was almost no fire in the first and fourth quarter, while both the

second and the third quarter disclose almost the same number of fire ignition incidents, 14854 and 17044 respectively. However when observing the forest area and non-forest area covered by fire, as well as the average forest fire expansion (for forest-area) the assumption is confirmed that the bulk of the yearly forest fires expansion happened in the third quarter 2010. In a robustness check we use the dataset provided by the Federal Forestry Agency (including the quarterly data) and successfully replicate our results. Figure 9: Forest fire statistics for the Russian forest fund on a quarterly basis for 2010 reported by the Federal Forestry Agency (data on a national level)

2,000

Forest area covered by fire (1,000 ha)

0

0

500

5,000

1,000

10,000

1,500

15,000

20,000

Forest fires

1Q

2Q

3Q

4Q

1Q

3Q

4Q

Average fire area per fire (ha)

0

0

100

50

200

100

300

150

400

Non-forest area covered by fire (1,000 ha)

2Q

1Q

2Q

3Q

4Q

1Q

2Q

3Q

4Q

12. Other robustness tests The empirical results of all robustness tests presented in this section including regression tables and graphical illustrations of the triple interactions can be found here. Table 10: Other robustness test Test Alternative dataset: Instead of using the forest fire statistics reported by the Russian Statistical Agency Rosstat we exploit a forest fire dataset compiled by the Federal Forestry Agency in order to replicate our results with yearly and quarterly data (see Appendix K on the differences between the datasets). Additional region-specific control variables: We control for a large set of region-specific characteristics

Additional governor-specific control variables: We control for a broad set of governor-specific characteristics

Bureaucratic and disaster management experience of governors: We test whether the governors have experience in managing a large number of bureaucrats and whether they were (potentially) confronted with crisis situations in the past.

Binary and continuous local origin variables:

Method (1) Replicate regressions using the new dataset only for wildfires on the forest area of the forest fund (2) Replicate regressions using the new dataset for wildfires on the forest and nonforest area of the forest fund (3) Use only quarterly data for the third quarter of 2010 (i.e. the period of extraordinary forest fires) – both only for forest area and for forest and non-forest area of the forest fund (for the distribution of forest fires on quarterly basis see Appendix K) Control for the following variables (if not otherwise specified, all variables are average over 2000-2009): (1) regional GDP as a measure of economic development, (2) fixed assets investments as a measure of economic development, (3) fiscal transfers (for all areas) as a measure of support from the central government (see note to the table), (4) road density (as a measure of forest accessibility for the urban population, contributing to faster wildfires expansion and/or easy access to wildfires by fire brigades), (5) crime rate (as a measure of general low obedience of the regional population contributing to both spread of fires and difficulty to mobilize volunteers to combat fires), (6) inflation-corrected growth rates of the regional GDP (which also serve as a proxy for the overall efficiency of the regional bureaucracy), (7) federal transfers for forest fire protection (reported in the Federal Forestry Agency dataset): we control for (7a) federal subsidies for the year 2010; (7b) federal subsidies for the third quarter of 2010; substitute regional forestry expenditures by yearly (7c) and 3rd quarter (7d) federal subsidies for forest fire prevention Control for the following variables: (1) professional background: dummies governors, who worked as businessmen, engineers and bureaucrats before their appointment (three most typical backgrounds), (2) dummy for governors with established business connections (e.g. businessmen in the past or owners of large companies), (3) dummy for members of United Russia, who may be receiving larger support from the center, (4) dummy for governors, who held a high-level position in the regional administration in the past (e.g. vice governor, member of the regional government, member of the regional parliament, mayor of the regional capital), (5) dummy for governors, who have never been subjected to public elections (i.e. came to power after 2004), (6) number of election victories (for governors, who were in office before 2004), (7) number of gubernatorial appointments under Putin and Medvedev, (8) age of the governor (as well as age interacted with log urban population and age squared) (9) tenure of the governor (as well as tenure interacted with log urban population and tenure squared) We set up measures of bureaucratic and disaster management experience of the governor: (1) Bureaucratic management experience, (2) Disaster management, (3) Exclude Bashkortostan region (because the governor worked in the Ministry of Emergency Situation) (4) Total fire expansion during governor tenure, (5) Average fire expansion during governor tenure (1) Use the binary index of local origin, i.e. a dummy variable equal to 0 for original

We use alternative approaches to constructing the index of local origin as opposed to what we have used in the main specification

Alternative dependent variables: We use alternative dependent variables capturing the impact of forest fires and regional forest fire crisis management Regions with zero forest fires: We test the assumption that regions with zero forest fires were speared from the natural disaster due to their topographic conditions Gubernatorial appointments/dismissals (summer months): In addition to the region excluded due to the ex ante model assumption, we exclude regions in which a governor had been appointed or dismissed in July and September in order to rule out confounding effects of gubernatorial appointments and dismissals Gubernatorial appointments (2010): We test whether our results are driven by a subgroup of governors which were appointed in 2010 (before the forest fire started) and potentially have less experience in regional governance Population variables: We test whether we are able to find consistent results if we use population, population, or mean-centered urban population instead of urban population. GDP: We test whether we are able to find consistent results if we use GDP instead of urban population (GDP may be a more appropriate measure determining the attention of the federal government than population – possibly, federal officials monitor primarily rich regions, and not the most populated ones) Former state-owned enterprise (SOE) managers: There are two governors in our sample which have a background as former managers in public enterprises. Due to the close links between politics and state owned companies it is necessary to test the effect of this small subgroup which has not been included in the federal connection dummy. Temporarily absent governors: Since the forest fires happened during the summer months in which Russian people, including governors leave their jobs for holidays. We check whether governors left for

local origin index of 1 and 2 and 1 for original local origin index of 3 and 4. Thereby we distinguish between governors who have spent a substantial part of their preinauguration life in the region and outsiders (regardless of whether the outsiders come from far or close regions) (2) Use a continuous local origin variable measuring the number of years the governor spent in the region before his inauguration (3) Use a continuous variable measuring the number of years the governor spent in the region before the outbreak of forest fires (combination of local origin and tenure variable) For approaches (2) and (3) we find a significant effect if the governor spent more than 10/20 years in his region of office before inauguration/forest fires respectively (the findings are robust if we control for age of the governor) (1) Dep. var.: forest area covered by fire (log); we replicate our results additionally controlling for the number of fires (log) (2) Dep. var.: average forest fire expansion (the standard dependent variable without log) Include five regions which have been previously excluded due to the absence of forest fires (for which the dependent variable was zero): Tula, Kalmykia, Ingushetia, Kabardino-Balkaria, and Northern Ossetia Exclude three regions in which governors had been appointed or dismissed during the extended forest fire period (July, August, and September): Bashkortostan (July 15), Novosibirsk (September 9), and Kaliningrad (September 27). Note: There were three other dismissals during and after the forest fires which we already excluded in our ex ante model assumptions. First, the governor of Chuvashia who resigned in August. Second, Moscow City, which we have excluded due the absence of forests. Third, Kalmykia, which was speared from forest fires in 2010. Exclude 12 regions in which a new governor was appointed before the summer forest fires (Karelia, Komi, Volgograd, Rostov, Dagestan, Bashkortostan, Tatarstan, Orenburg, Chelyabinsk, Krasnoyarsk, Sakha, Evreyskaia).

Substitute log urban population with log population, log population density and meancentered urban population (including the respective interaction terms)

Substitute log urban population with log GDP (in interaction terms as well)

(1) Control for dummy which is 1 for the regions which are governors by former SOE managers (Samara and Saratov) (2) Compute a new “version” of our federal connection dummy including former managers of public enterprises and replicate the results using this dummy

We were able to identify 6 regions in our sample in which the governor was on holiday leave around mid July (based on media reports): Vladimir, Voronezh, Udmurtia, Kirov, Samara, and Sverdlovsk. Due to public protest in the regional media and pressure from the federal center (personal criticisms from Putin himself) all of the absent governors urgently interrupted their holidays and actively participated in the

their holidays (by screening newspaper articles verifying the participation of governors at catastrophe management).

forest fire combat activities. We control for a dummy which is 1 for governors who were on holiday but returned by end of July, otherwise 0.

Regional parliamentary elections: In 6 regions parliamentary elections were taking place in October 2010 (Magadan, Tyva, Novosibirsk, Chelabinsk, Kostroma, and Belgorod). The elections were scheduled before the forest fires and could have an influence on the governors’ behavior. We check whether these regions are driving our results. Regions with federal emergency situation: In July 2010 President Medvedev proclaimed the federal state of emergency in several regions. We check whether our results are driven by these regions. Outliers: We check for the influence of outliers

Control for a dummy which is equal to 1 if there was a regional parliamentary election in the region in October 2010, otherwise 0.

Local origin of vice-governors: We test whether our results are driven by the local origin of vice-governors instead of governors Weather conditions: In a number of regressions we control for actual temperatures and rainfall in July 2010, temperature anomalies in July 2010, as well as yearly average wind velocity.

Federal districts: We test for the regional affiliation the effect of federal districts in order to control for climatic and geographic characteristics. Types of trees: We control for the most prevalent tree types and their respective wood characteristics which could influence the velocity with which wildfires can spread. Spatial autocorrelation: We control for possible spatial interdependence between forest fires in neighboring regions “Patristism” of local elites: We test for region- and governor –specific variables capturing patriotism of local elites

Effectiveness of local bureaucracy: We control for variables measuring the size and effectiveness of the regional bureaucracy

Control for a for dummy which is equal to 1 if the region received federal emergency status (Mari El, Mordovia, Vladimir, Voronezh, Moscow region, Nizhny Novgorod), otherwise 0

(1) Calculate Cook`s distance values for each region. Use cut-off point of 4/n, with n being size of the sample: Voronezh (0.465), Altai Rep. (0.291), Sverdlovsk (0.152), and Stavropol (0.076) have distance values above the cut-off rate. Exclude each region separately and test for the robustness of the previous findings. (2) Exclude each of the remaining regions separately and test for the robustness of the previous findings We compile a dummy equal 1 for regions with vice-governors of local origin and zero otherwise (for more details see Appendix 12 of the Supplementary Materials).

(1) Substitute the actual temperature and rainfall in July 2010 for the long term temperature and rainfall in July (2) Re-estimate our baseline regressions while additionally controlling for temperature anomalies in July (anomalies are positive and significant at the 5% level) (3) Re-estimate our baseline regression while substituting temperature anomalies for the long-term temperature in July 2010 (anomalies are positive and significant at the 5% level) (4) Re-estimate the baseline regressions while additionally controlling for average yearly wind velocity (wind velocity is positive but insignificant) Control for federal district dummies

Control for 13 variables capturing the most common forest tree species in Russia measure by the area covered by a particular type of tree in a particular region (alder, ash, aspen, beech, birch, cedar, fir, larch, lime, maple, pine, spruce, and oak).

Estimate spatial lag and spatial error regressions, using two types of spatial weights matrices: binary matrix for border regions and inverse railroad travel distance matrix by Abramov (2008) Region-specific variables capturing the patriotism of local elites: (1) ethnic republics, (2) Northern Caucasus republics, (3) share of ethnic Russian population, (4) ethno-linguistic fractionalization, (5) polarization, (6) distance from Moscow, (7) oil and gas extraction Governor-specific variables capturing the patriotism of local elites: (8) tenure (predecessor governor), (9) number of predecessors Proxies for size and effectiveness of regional bureaucracy including: (1) share of bureaucrats with university degree, (2) professional development educational program, (3)-(4) index of efficiency and index of improvement of efficiency (both compiled by the (Ministry of Regional Development), (5) size of bureaucracy, (6) number of agencies, (7) number of municipalities, (8)-(10) public sector efficiency indicators, (11) index of sub-national democracy, (12) index of

freedom of press, (13) experienced corruption, (14) perceived corruption, (15) tax revenue / GDP, (16) income tax revenue / GDP, (17) tax arrears / GDP, (18) crime rate, (19) relative extractive capacity, (20) infant mortality

Quality of human capital, intrinsic motivation and connection of governors: We control for other proxies of local knowledge such as education, intrinsic motivation and local relationship

Alternative specifications

Proxies for effectiveness of forest management in the region: (21)-(24) Quality of forest governance: organization, law, economic efficiency, ecological performance, (25) no independent forest agency, (26) average yearly fire expansion 1992-2009 In order to capture the level of education we control for the following variables: (1) web of world universities ranking, (2) HSE ranking, (3) Higher education in Moscow or St. Petersburg, (4) studied outside his region of birth, (5) mobility For intrinsic motivation and local connection we test the following variables: (6) non-ethnic Russian governor, (7) governor ethnicity corresponds to titular nation of the region, (8) rural origin, (9) involvement of local interest in forest governance, (10) governor was born in the region of office (1) Replicate baseline regression including all covariates not orthogonal to local origin or federal connections variables, extracted from all robustness checks we used (2) Estimate regressions without any control variables

Note: Fiscal transfers are financial contributions transferred from the federal center to the regional government according to an equalization formula based to the socio-economic situation of the region. We consider the average fiscal transfers for the period 2000-2009. The use of this variable is due to our attempt to capture the long-term financing of the region by the federal government. For 2010, looking at the transfers would be insufficient, since a big portion of aid received by the region was not captured as transfers (e.g. various federal assistance programs, with partly intransparent financial flows) or happened not in form of transfers (e.g. use of fire planes and military resources), hence, this variable would not be representative for our study. In a robustness check, we do control for federal transfers for forestry in 2010, and still find a significant effect of our variables; it indicates that this “non-financial” or “hidden” part of the federal support is particularly important. On the discretionary nature of Russian federal fiscal transfers see Popov (2004); Jarocinska (2010); Yakovlev et al. (2011).

13. Definitions of the variables Table 11: Definitions of the variables Variable Description Age Total years of life of the governor from his birth to 2010 Average fire Yearly average forest area covered by fire during the governors` tenure. expansion during governors` tenure Average fire expansion during the period 1992-2009 Average spread of a forest fire

Born in the region of office Bureaucracy Bureaucrat Bureaucratic management Business connection

Businessman Crime Democracy

Disaster management Distance from Moscow Dummy no independent forest agency Education in Moscow / St. Petersburg Education not in region of birth Efficiency of bureaucracy

Yearly average forest area covered by fire during the period 1992-2009.

Log (Area covered by forest fires, hectare / Number of reported forest fires + 1), 2010: a) Rosstat data b) Federal forestry data (2010: forest area + forest and non-forest area) c) Federal forestry data (3Q 2010: forest area + forest and non-forest area) Dummy variables: 1 for governors who have been born in the region of office, otherwise 0 Log number of bureaucrats in the executive branch (log transformation is used to eliminate outliers), 2010 Dummy variable: 1 for governors who commenced a career in public administration and politics after graduation, otherwise 0 Total number of years the governor used to manage a large number of bureaucrats until the forest fires in 2010. Dummy variable: 1 for governors who have close ties to companies, have been entrepreneurs in the past, or are known to hold majority share in big companies, otherwise 0 Dummy variable: 1 for governors who commenced a business career after graduation, otherwise 0 Number of crimes per 100,000 inhabitants, 2010 Expert opinion indicator of the level of sub-national democracy in Russian regions in 2000-2004, computed as sum of 10 expert opinion assessments of the following dimensions of regional political life: political openness, freedom of elections, political pluralism, freedom of media, economic liberalization, development of civil society, regional political organization, balance of power in regional elites, political corruption and independence of municipalities, each evaluated on 5-point scale with 5 being the highest value. Thus, increasing values of the democracy index indicate higher value of sub-national democracy, with 50 being the highest possible indicator Dummy variable: 1 for a governors with work experience in the federal Ministry of Emergency Situations and two governors with military combat experience, otherwise 0 Geographical distance between Moscow and the capital of the region (`000 km), zero for Moscow region Dummy equal to 1 for regions which have no independent forest agency in their administration (data is unfortunately available only for 2013)

Source Own calculation Federal Forestry Agency (reported by Rosstat) and own calculation Federal Forestry Agency (reported by Rosstat) Two datasets: Rosstat and Federal Forestry Agency (reported by Rosstat) Own calculation Rosstat Own calculation Own calculation Own calculation

Own calculation Rosstat Moscow Carnegie Center

Own calculation

Rosstat Russian Forest Agency website

Dummy variable, 1 if governor graduated from a university in Moscow or St. Petersburg, 0 otherwise

Own calculation

Dummy variable, 1 if governor graduated from a university located not in the region of birth, 0 otherwise Four indices of quality of sub-national bureaucracy based on analysis of its performance and measuring: (a) extent of application of objective-oriented public management (b) transparency and incentives in the internal organization of bureaucracy (c) bureaucratic practices in interaction with recipients of public services Throughout the paper, these indices are denoted as indices I, II and III respectively

Own calculation Aleskerov et al. (2006)

Efficiency of regional government

Engineer Ethnic fractionalization Experienced corruption

Federal connections

Federal districts Federal emergency Federal subsidies for forest fire protection

Fire expansion during governors` tenure Fiscal transfers

Forest area Forest area covered by fire Forest fire incidents Forestry expenditures Freedom of the press

GDP GDP growth rate Improvement of efficiency of regional government

Income tax revenue to GDP

Original index was assigned values A+, A, B or C, in the current paper we set its value from 1 to 4, with 1 being the highest level of efficiency Ranking computed by the Ministry of Regional Development of Russia measuring the efficiency of regional government, based on weighting of various aggregate economic indicators of the region. The regions are divided into four groups according to their place in the ranking, first three including 20 regions each, and the last 23 regions. The resulting indicator varies from 1 to 4, with 1 being the highest level of efficiency Dummy variable: 1 for governors which commenced a career in engineering after graduation, otherwise 0 Index of ethnic fractionalization, computed based on Census of 2002 data using the approach in Alesina and La Ferrara (2005) Index which is based on 2010 public opinion survey which varies between 0 and 1, of which 1 is the highest possible corruption level. The index measures the experienced corruption: how often individual report to have been requested and have paid a bribe Dummy variable: 1 for governors, who have worked in the federal administration as ministers, officials of presidential administration, high-ranked officials in the ministries or presidential representatives in various regions, 0 otherwise; membership in federal parliament does not count as federal connections Set of dummies for federal districts: Central, Northwestern, Siberian, Southern, Northern Caucasus, Far East, Volga and Ural Dummy variable, 1 for regions where president Medvedev proclaimed the status of emergency in July 2010, 0 otherwise Subvention provided by the federal budget specifically for the security of forests from wildfires: construction (and re-construction and maintenance) of roads, airfields, aircrafts and helicopters with firefighting importance, the implementation and maintenance of early-warning fires systems, the formation of storage of fuel and lubrication materials, forest fire monitoring (ground, air, satellite), forest fire extinction (ground and air) and other fire safety measure (e.g. construction of fire preventing “mineral band / barriers”, preventive and prophylactic burning off combustive materials, installation of drive-way access gates, publications and appearance in mass media etc.), (‘000 RUB), 3Q 2010 and 1Y 2010. Note that in the 4Q 2010 five regions reported negative subsidies. The sum of yearly forest area covered by fire starting from the year of gubernatorial appointment until 2009

Share of transfers from all other budgets, as well as (partly) state-owned corporations (gosudarstvennaya korporatsiya) in regional expenditures, average for 2000-2009 Share of regional territory covered by forest, %, 2009 Forest area covered by fire (hectares) Number of forest fires (incidents) Regional expenditures on forestry (‘000 RUB), average for 2000-2009 Index of the freedom of the press, measuring the freedom of access to, production of and distribution of information, higher values represent higher level of press freedom Log gross regional product per capita (RUR), average for 2000-2009 Annual GDP growth rate (inflation corrected), %, average for 2000-2009 Ranking computed by the Ministry of Regional Development, measuring the improvement of the efficiency of regional government in 2007-2009, based on weighting of various aggregate economic indicators of the region. The regions are divided into four groups according to their place in the ranking, first three including 20 regions each, and the last 23 regions. The resulting indicator varies from 1 to 4, with 1 being the highest level of efficiency Share of income tax revenue of the regional budget in the regional GDP, 2010 (since income tax is distributed between different levels of budgetary system in

Ministry of Regional Development

Own calculation National Census FOM

Own calculation

Own calculation Own calculation Federal Forestry Agency (reported by Rosstat)

Federal Forestry Agency (reported by Rosstat) and own calculation Ministry of Finance; State Treasury Rosstat Rosstat Rosstat Ministry of Finance; State Treasury Institute of Public Expertise Rosstat Rosstat Ministry of Regional Development

Rosstat

Infant mortality Investment Involvement of nongovernmental actors in forest governance

Local origin

Mobility Never elected Non ethnic Russian governors Northern Caucasus

Number of appointments Number of electoral victories Number of municipalities Number of predecessors Oil and gas Perceived corruption

Polarization Population Population density Professional training of bureaucrats Quality of governance in forest sector

Rainfall July Regional office

fixed proportion, using this proxy is equivalent to using income tax revenue of all budgets of Russia) Number of infants, who died at an age of 1 year or younger, per 1,000 born infants, 2010 Fixed capital investments, averaged for 2000-2009 (million RUB) Index measuring the degree of involvement of regional non-governmental actors in the forest governance The original index was assigned five values: A1, A2, B1, B2 and C and is mostly based on aggregation of statistical indicators. For this paper, we recomputed the index on a scale from 1 to 5 with 1 being the highest quality of governance (a) Baseline index: index varying from 1 to 4 with 4 being the highest level of local origin (governor spent most of the pre-inauguration life in the region he rules) and 1 being the lowest level (governor spent most of the pre-inauguration life in a region distant from the region he rules) (b) Binary index: dummy variable, with 1 being a governor with local origin (values of baseline index 3 or 4) and 0 being a governor without local origin (values of baseline index 1 or 2) (c) Number of years spent by the governor in his region before inauguration (d) Number of years spent by the governor in his region before the commencement of forest fires Number of regions the governor worked after graduation up to his appointment Dummy variable: 1 for governors who never have been elected and consequently only have been appointed, otherwise 0 Dummy variable, 1 if governors are not ethnic Russians, 0 otherwise

Rosstat Rosstat WWF

Own calculation

Own calculation Own calculation Own calculation

Dummy variable: 1 if the region is a republic located in the Northern Caucasus (Adygeya, Kabardino-Balkaria, Karachaevo-Cherkessia, Northern Ossetia, Ingushetia and Dagestan), 0 otherwise Number of times governor was appointed or re-appointed into his office by either Putin or Medvedev (in Russia until 2010 there have been no individuals, who consequently occupied gubernatorial positions in several regions) Number of times governor was publicly elected into his office in his region (in Russia until 2010 there have been no individuals, who consequently occupied gubernatorial positions in several regions) Total number of municipalities in the region, 2010

Own calculation

Number of governors preceding the current governor as heads of the region (after the collapse of the USSR) Oil and gas extraction in the region in coal equivalents (mln t oil * 1.4 + bln cubic m gas *1.2), average for 2000-2007 Index which is based on 2010 public opinion survey which varies between 0 and 1, of which 1 is the highest possible corruption level. The index measures the perception of corruption in the region Index of polarization of ethnic structure, computed based on Census of 2002 data using the approach in Alesina and La Ferrara (2005) Log population, end of year, people, average for 2000-2009 Log (Population, end of year / area of the region as of 2009), average for 20002009 Staff received additional professional training / Total staff of state agencies and municipalities, average for 2008-2009 Four indicators measuring the quality of governance in the forest sector of the region, in particular (a) organization of forest governance (size of respective bureaucracy, its efficiency etc.); (b) legal framework for forest governance and its implementation; (c) performance of forest governance in terms of environmental protection and (d) performance of forest governance in terms of economic efficiency of forest use. The original indices were assigned five values: A1, A2, B1, B2 and C and is mostly based on aggregation of statistical indicators. For this paper, we recomputed the indices on a scale from 1 to 5 with 1 being the highest quality of governance Average long-term rainfall in July (mm); rainfall in July 2010 (mm) Dummy variable: 1 for governors who previously held a high-rank position in

Own calculation

Own calculation

Own calculation

Rosstat

Rosstat FOM

National Census Rosstat Rosstat Rosstat WWF

Rosstat Own calculation

Regional parliamentary elections Relative extractive capacity

Republic

Road density Rural origin

Same ethnicity

Share of bureaucrats with university education Share of ethnic Russians State-owned enterprise manager dummy Tax arrears to GDP Tax revenue to GDP Temperature anomalies July 2010 Temperature July Temporary absent governors Tenure Tenure of predecessor Territory Top 30 universities, HSE ranking Top 30 universities, Ranking Web of World Universities Type of trees

United Russia Urban population Vice-governors local origin Wind velocity

the regional government, otherwise 0 Dummy variable: 1 for regions where regional parliamentary elections took place in October 2010, 0 otherwise The index is computed in the following way. First, we estimate a regression, where tax revenue to GDP is the dependent variable, and GDP, share of mining in GDP, share of exports in GDP and health expenditures to GDP are explanatory variables. Then we compute counterfactual level from the tax revenue to GDP from this regression. Finally, we divide the actual tax revenue to GDP over the counterfactual value Dummy variable: 1 if the region has the status of an ethnic republic (a subgroup of Russian regions with typically relatively high share of non-Russian population, which in the past enjoyed higher level of autonomy and still have strong differences from the center in terms of composition of regional elites and political practices), 0 otherwise Length of hard roads (km) per 1,000 km of the regional territory, 2000-2009 Dummy for governors, who have spent an important part of their career working in the agricultural sector or forestry or in organizations governing agriculture and forestry (including Duma committees in this area) Dummy variable, 1 if governor in an ethnic republic has the same ethnicity as the titular ethnicity of the republic (i.e. the ethnicity, which provides the republic its designation, e.g. Tatars in Tatarstan etc.; while this ethnic group does no necessarily constitute a majority in the republic, it typically has very strong positions in the republican elites), 0 otherwise Share of civil service officials that graduated from a university or an equivalent educational establishment, as of 1st October 2009

Own calculation

Rosstat

Own calculation

Rosstat Various websites

Own calculation

Rosstat

Share of ethnic Russians in the regional population, 2002

National Census

Dummy variable: 1 for governors, who have been managers of big state-owned companies, 0 otherwise

Own calculation

Share of tax arrears in the regional GDP, 2010 Share of total tax revenue (of regional, federal and municipal budgets) from the territory of the region in the regional GDP, 2010 Actual average temperature in July 2010 minus long-term average temperature in July Average long-term temperature in July (Celsius); temperature in July 2010 (Celsius) Dummy variable, 1 for governors, who were temporary absent (on vacation) during the early forest fire period, according to media reports, 0 otherwise Number of days the governor stays in office from inauguration as governor (president) until July 21st , 2010 Number of days the predecessor of the current governor stayed in office from inauguration until the resignation Log territory, ‘000 km2 Dummy variable, 1 if governor graduated from one of top 30 universities in the Ranking of the Higher School of Economics (HSE), 0 otherwise Dummy variable, 1 if governor graduated from one of top 30 universities in the Ranking Web of World Universities, 0 otherwise

Rosstat Rosstat

Major forest tree species in Russian regions as of 2003 (alder, ash, aspen, beech, birch, cedar, fir, larch, lime, maple, pine, spruce and oak), in 103 hectares Dummy variable: 1 for governors, who are members of United Russia, 0 otherwise Log urban population, people, average for 2000-2009 Dummy variable: 1 for vice-governors with local origin, 0 otherwise Average yearly wind velocity (meter per second at 10 meter altitude); mean of

Rosstat Rosstat Own calculation Own calculation Own calculation Rosstat Own calculation Own calculation

Federal Forestry Agency Own calculation Rosstat Own calculation Energy Wind

the average yearly wind velocity (measured by all regional wind stations) divided by the number of regional wind stations (number of regional wind stations vary between 3 and 43) Note: (a) Unfortunately Rosstat only reports the annual data for actual average temperature and rainfall in July and the respective deviations from the norm. The norm represents the average temperature and rainfall in July for the reference period (1960-1989). In order to obtain the norm, or in other words the long-term average, we add (or subtract) the deviations from the norm from the actual reported temperature and rainfall in July 2010 (b) A few regions did not report any public expenditure on forestry for some years. (c) Rosstat ceased to report regional oil and gas extraction after 2007 (d) Rosstat refers to the Federal Statistical Agency of the Russian Federation (e) “University” includes any recognized university-rank school, regardless of designation (e.g. institutes, academies etc.) (f) The dataset for the regional occurrence of specific tree species can be downloaded from http://webarchive.iiasa.ac.at/Research/FOR/forest_cdrom/english/for_fund_en.html

14. References Abramov, A. (2008). The shortest railway distance between capital cities of Russian regions. Mimeo. Acemoglu, D., Garcia-Jimeno, C. & Robinson, J.A. (2014) State Capacity and Economic Development: A Network Approach. Mimeo Alesina, A., & La Ferrara, E. (2005). Ethnic diversity and economic performance. Journal of Economic Perspectives, 43, 762-800. Aleskerov, F., Golovishchenskiy, K., & Klimenko, A. (2006). Otsenki Kachestva Gosudarstvennogo Upravleniya. HSE Working Paper No. WP8/2006/02. Arikan, G. G. (2004). Fiscal decentralization: a remedy for corruption? International Tax and Public Finance, 11, 175-195. Baes, J.E. & Santos, I.V. (2007). Children’s Vulnerability to Weather Shocks: A Natural Disaster as a Natural Experiment. Mimeo. Binz-Scharf, M. C., Lazer, D., & Mergel, I. (2012) Searching for answers: networks of practice among public administrators. American Review of Public Administration, 42, 202-225. Blokov, I. (2010). Alle Jahre wieder: Waldbrände in Russland. Osteuropa, 9, 5-16. Bracanti, D. (2007). Political aftershocks: the impact of earthquakes on intrastate conflicts. Journal of Conflict Resolution, 51, 715-743. Brown, J. D., Earle, J. S., & Gehlbach, S. (2009). Helping hand or grabbing hand? State bureaucracy and privatization effectiveness. American Political Science Review, 103, 264-283. Brückner, M., & Ciccone, A. (2011). Rain and the democratic window of opportunity. Econometrica,79, 923-947. Burke, P. J., & Leigh, A. (2010). Do output contractions trigger democratic change? American Economic Journal: Macroeconomics, 2, 124-157. Cheibub, J. A. (1998). Political regimes and the extractive capacity of governments: taxation in democracies and dictatorships. World Politics, 50, 349-376. Dininio, P., & Orttung, R. (2005). Explaining patterns of corruption in the Russian regions. World Politics, 57, 500-529. Fishman, R., & Gatti, R. (2002). Decentralization and corruption: evidence across countries. Journal of Public Economics, 83, 325-345. Goldsmith, A. A. (1999). Slapping the grasping hand: correlates of political corruption in emerging markets. American Journal of Economics and Sociology, 58, 865-886. Grossman, P. J. (1994). A political theory of intergovernmental transfers. Public Choice, 78, 295-303. Gurgur, T., & Shah, A. (2005). Localization and corruption: panacea or pandora box? World Bank Policy Research Working Paper No. 3486. Hanson, J. K., & Sigman, R. (2013). Leviathan’s latent dimensions: measuring state capacity for comparative political research. Mimeo. Healy, A., & Malhotra, N. (2010). Random events, economic losses, and retrospective voting: implications for democratic competence. Quarterly Journal of Political Science, 5, 193208. Hendix, C. S. (2010). Measuring state capacity: theoretical and empirical implications for the study of civil conflict. Journal of Peace Research, 47, 273-285. Hernandez, M. A., Robles, M., & Torero, M. (2010). Fires in Russia, wheat production, and volatile markets: reasons to panic? IFPRI Report, August 6. Hoff, K. &. Stiglitz, J. E. (2004). After the big bang? Obstacles to the emergence of the rule of law in post-communist societies. American Economic Review, 94, 753-763.

Idrisova, V., & Freinkman, L. (2010). Vliyanie Federal’nykh Transfertov na Fiskal’noe Povedenie Regional’nykh Vlastei. Moscow: IEPP. Jarocinska, E. (2010). Intergovernmental grants in Russia: vote-buying or bargaining power of the regions. Economics of Transition, 18, 405-427. Khemani, S. (2004). Partisan politics and intergovernmental transfers in India. World Bank Policy Research Working Paper No. 3016. Kugler, J., & Arbetman, M. (1997). Relative political capacity: political extraction and economic research. In M. Arbetman & J. Kugler (Eds.), Political capacity and economic behavior (pp. 11-46). Boulder: Westview Press. Libman, A. (2012). Democracy, size of bureaucracy, and economic growth: evidence from Russian regions. Empirical Economics, 43, 1321-1352. Libman, A., & Feld, L. P. (2013). Strategic tax collection and fiscal decentralization: the case of Russia. German Economic Review, 14, 449-482. Libman, A., Kozlov, V., & Schultz, A. (2012). Roving bandit in action: outside option and government predation in autocracies, Kyklos, 65, 526-562. Lieberman, E. S. (2002). Taxation data as indicators of state-society relations: possibilities and pitfalls in cross-national research. Studies in Comparative International Development, 36, 89-115. Miguel, E., Satyanath, S., & Sergenti E. (2004). Economic shocks and civil conflicts: an instrumental variables approach. Journal of Political Economy, 112, 725-753. Orttung, R. (2004). Business and politics in the Russian regions. Problems of PostCommunism, 51, 48-60. Ponomareva, M., & Zhuravskaya, E. (2004). Federal tax arrears in Russia: liquidity problems, federal redistribution or regional resistance? Economics of Transition, 12, 373-398. Popov, V. (2004). Fiscal federalism in Russia: rules versus electoral policies. Comparative Economic Studies, 44, 515-541. Porfiriev, B. (2012). Economic issues of disaster and disaster risk reduction policies: international vs. Russian perspectives. International Journal of Disaster Risk Reduction, 1, 55-61. Prud’homme, R. (1995). The dangers of decentralization. World Bank Research Observer, 10, 201-220. Reinikka, R., & Svensson, J. (2004). Local capture: evidence from a central government transfer program in Uganda. Quarterly Journal of Economics, 119, 679-705. Ross, C. (2010). Federalism and inter-governmental relations in Russia. Journal of Communist Studies and Transition Politics, 26, 165-187. Russian Institute of Global Climate and Ecology (2010a) Climate Change 2010. Spring Report Russian Institute of Global Climate and Ecology (2010b) Climate Change 2010. Summer Report Sala-i-Martin, X. (1997). I just ran two million regressions. American Economic Review, 87, 178-183. Schedler, A.(2009) Electoral authoritarianism. In T. Landman & N. Robinson (Eds.) The Sage Handbook of Comparative Politics (pp. 381-394). London: Sage. Schultz, A., & Kozlov, V., & Libman, A. (2014). Judicial alignment and criminal justice: evidence from Russian courts. Post-Soviet Affairs, 30, 137-170. Simpser, A. (2013). Why governments and parties manipulate elections? New York: Cambridge University Press Slinko, I., Yakovlev, E., & Zhuravskaya E. (2005). Laws for sale: evidence from Russia. American Law and Economics Review, 7, 284-318. Soares, M. M., & Neiva, P. R. P. (2011) Federalism and public resources in Brazil: federal discretionary transfers to states. Brazilian Political Science Review, 2, 94-116.

Steinmo, S. H. (1993). Taxation and democracy. New Haven: Yale University Press. Tanzi, V. (2000). Some politically incorrect remarks on decentralization and public finance. In J. J. Dethier (Ed.), Governance, decentralization and reform in China, India and Russia (pp. 47-63). Boston: Kluwer. Treisman, D. (2000). The causes of corruption: a cross-national study. Journal of Public Economics, 76, 399-457. Treisman, D. S. (1999). After the deluge: regional crises and political consolidation in Russia. Ann Arbor: The University of Michigan Press. Treisman, D. S. (2003). Fiscal pathologies and federal politics: understanding tax arrears in Russia’s regions. In J. McLaren (Ed.) Institutional elements of tax design and reform (pp. 49-68). Washington: World Bank. Turovskii, R. (2010). How Russian governors are appointed: inertia and radicalism in central policy. Russian Politics and Law, 48, 58-79. Wedel, J. R. (2003). Clans, cliques and captured states: rethinking “transition” in Central and Eastern Europe and the former Soviet Union. Journal of International Development, 15, 427-440. Yakovlev, A., Marques, I., & Nazrulaeva, E. (2011). From competition to dominance: political determinants of federal transfers in the Russian Federation. Mimeo.