Seasonality in local food markets and household consumption in Africa: Evidence from 3 ESA countries

J. Kaminski, L. Christiaensen, C. Gilbert, and C. Udry

AG RICULTURE

IN AFRICA T E L L I N G FA C T S FROM MYTHS

This paper provides two new contributions • New evidence on the seasonal patterns of agricultural markets and household food and non-food consumption in SSA – Joint analysis of long market time series and several nationally-representative household surveys (implemented over more than 12 consecutive months) – Systematic measurement and identification

• New measures consistently defined and applied for both time-series and survey data analysis and relationship between market and household-level seasonality • Seasonality appears to be high and still prevalent for most locations and crops, with significant consequences for household expenditures • Towards a renewed interest given newly-available data for Africa? Page 2

What is seasonality and how does it come about? • Seasonality: – Regular intra-annual variability at predetermined time periods/months, i.e. “the deterministic part of intra-annual volatility” = a major component of overall volatility in Africa – Repeated patterns within a year can remain constant or change over time

• Regular drivers of seasonality in prices – Seasonality in production – production cycles/AEZs – Seasonality in demand (e.g. Holidays, school calendar and fees...)

Some minimal seasonality in prices is expected given storage and capital costs • Others drivers – Role of marketing & transaction costs (limited spatial arbitrage) – Market structure in the wholesale, comparative advantages in storage – Marketing behavior of farmers in response to liquidity constraints (sell low, buy high) and credit constraints, savings vehicles (Deaton, Osborne)

• Why does it translate in seasonality in consumption (food and non-food)? – Lack of non-seasonal or counter-seasonal income-generating activities (e.g. migration) – Lack of other coping mechanisms (self and community insurance) and savings – Depends on degree of food-non food substitution elasticity and absolute price levels Page 3

Conventional wisdoms • Seasonality in prices has been much studied and was considered to have substantial welfare effects – Sahn, 1989; Reardon, 1997; Dorward et al 2004

– Eg severe acute malnutrition in Niger –WHZ- bw 1% -4% (harvest-hungry seasons) (ACF, Araujo-Bonjean)

• Seemingly considered less important today? – Agricultural markets are more integrated today than in the nineties – Little studied/discussed during the past 15 years (except Devereux 2009) – Poverty studies often do not correct for it – some recent studies do and find significant effects (e.g. Dercon & Krishnan 2000; Khandker)

• Still an issue? – Markets are well integrated within countries, but much less across borders; and subject to significant amount of volatility (after liberalization policies of the 90s2000s) : seasonality is a key ingredient of volatility in SSA – Also market integration does not exclude price seasonality if correlated production cycles across space – Many studies only covering certain markets in certain areas (publication bias) – Renewed interest (Devereux, 2009)? – Exploring both price and consumption seasonality more systematically Page 4

Our contribution • Rigorous discussion and construction of seasonality measures – Depart from the development micro literature (abstracting from time series analysis / cursory treatment) – Unconditional & Conditional – Bring in insights from finance (market time series) and public health literature (seasonality as an inequality concept)

• Updated and enlarged evidence basis (across countries and regions) – Three ESA countries (UG, TZ, MWI), 2 waves for TZ: internal and external validity – Seasonality measures of agricultural monthly prices series (2000-2013) in 869 markets in total – Seasonality measures of both food and non-food consumption in the same countries, derivation of consumption cycles

• Link food market price & household consumption behavior (juxtaposition and correlation) + disaggregated / heterogeneity analyses Page 5

Our data • Market prices: construction of 13 years of monthly-based time series – Staples, some fruits and vegetables + pulses, some processed products – Both wholesale and retail market levels – For the main consumption centers in each country (868 markets in total) – Imputation of missing values

• LSMS-ISA data at the household level (TZ 2008, MWI 2010, UG 2009 ) – Around 3,000 hhs in UG and TZ; 12,000 in MWI – Cross-section data but collected over more than 13 months in each region with about equal sub-sample slices in each month (no repeated observations over time) – Availability of some longitudinal data (full panels to be released) – Food and non-food expenditure data available (spatially and intra-annually deflated) as well as rich set of co-variates (hh income and assets, education, size, labor market status of the hh head, AEZ…), consistent construction of consumption aggregates (CLSP methodology) in terms of imputation and deflation, interpretable in quantities Page 6

Our findings Seasonality in food prices & expenditures remains pervasive • 25% seasonal mark up in maize price; abt 10-15% for staple food price index • 40-45 % seasonal mark up in food consumption and an even larger 65-70% seasonal mark up in non-food consumption • Strong heterogeneity across locations and sub-samples between and within countries (e.g. urban vs. rural households) • Price and expenditure seasonality seem linked in the annual cycles – Food and non-food cons seasonal patterns inversely follow staple price seasonal patterns but heterogeneously across and within countries – Uganda cons. seasonality =3* price seasonality measures but Malawi is eq. to 1/3: huge income effects but year-exceptional and non-price factors are also captured in the cons. estimates – e.g. in Tanzania: irregular and exceptional seasonality in 2008-9

• Policy implications – Understanding excess price seasonality – Intra-annual consumption smoothing remains a challenge – More attention to intra-annual timing of surveys needed in poverty measurement and analysis Page 7

Seasonality measures • Seasonality measures – Based on seasonal (monthly) factors – A gap measure between the extreme month effects – a range value for the amplitude of the seasonal cycle – Intra-annual standard deviation for measuring the extent of intra-annual variability – A gini as an overall inequality measure

• Unconditional/descriptive vs conditional/estimated – Unconditional/descriptive: based on the average month values of the sample – as in the development literature (for market prices), often for specific years (e.g. food security crises) though – we rather look at longer time series to detect regular seasonality – Conditional/estimated: based on seasonal factor estimates after modelling (accounting for trend & autocorrelation in times series, or sample heterogeneity in survey data)

Page 8

Unconditional/descriptive measures • Seasonal effects 1 n sm    pym  p  n y 1

p

1 n 12  pyj 12n y 1 j 1

• Gap and ssd 1 12 2 ssd  sm  11 m1

gap  max  sm   min sm 

• Gini  mC 1  s

 13 1 13 1 Gini     m 1  s      ms  12 72 12 72 6  C 1  s   12

m1 12

m1

m

12

m 1

12

m

m 1

m

m

Page 9

Empirical methodology : estimations of price time series – Price time series: a flexible de-trending ARIMA model with multi-cyclical (eg Arusha) and time-varying seasonality specification k

ln pmt    mj ln pmt  j  mt  smt   mt j 1

mt  mt 1  mt   mt mt  mt 1  mt 6

smt    jmt j 1

 jt    jt   where  jmt   jm cos    sin  jm    6   6 

– General trend model – deterministic or stochastic: random walk with a stochastic drift – Estimated parameters are the three innovation terms – Autoregressive lags to ensure serial independence – A trigonometric representation of seasonality which better accomodates for timevariability of the seasonal patterns   jm   jm   cos  sin        jm  6 6      j ,m1    jt   *     *    *    jm  jm       jm   sin  j ,m1   jt    cos     6   6  

 j  1,

,6 

Page 10

Empirical methodology : estimations of hh consumption – OLS regression with household and geographic controls + month dummies

ln c itf / nf  β f / nf X i  Mtf / nf  sitf / nf   itf / nf

o Specification issues on survey weights and strata and regional controls – use several specifications as robustness checks for the seasonality measures – first with the time-invariant hh characteristics only, and then control for possibly seasonal characteristics (income, assets) – Objective (i) is to control for systematic differences across the monthly slices of the sample across households that are not due to seasonality per se but to survey implementation (e.g. urban bias during rainy season?) – Yields a high range of seasonality measure (driven by price, income, and other) – Additional RHS help robustness and improve objective (i) but subject to recall bias or capture some of the LHS seasonality : try several RHS specifications but focus on the base one.

o Applied to national and sub-national samples to derive aggregate, disaggregate, and reaggregated conditional seasonal factors – the month dummy coefficients – So as to deal with the aggregation/attenuation bias driven by heterogeneous timing of seasonality – Specification tests show that re-aggregated indicators and estimates are much more robust and of higher levels, indicating attenuation bias Page 11

Empirical methodology: conditional seasonality measures • Price and expenditure gaps (weak principle of transfer) gm  max{ smt }  min{ smt }  2   m2  m2 (when the 6-month cycle predominates)

• Expenditure gap G f / nf  max{Mtf / nf }  min{Mtf / nf }

• Seasonal gini in expenditures = seasonal inequality of consumption levels – To what extent would seasonal inequality be reduced, on average over time, by a transfer of consumption from a high to a low month?” (principle of transfer) – Can be approximated as equal to 0.2G when the 6-month cycle predominates 12

Gini 

 tC 1  M  t 1 12

t

6  C 1  Mt 

13 1 12 13 1 12    t 1  Mt     t 1  Mt  12 72 t 1 12 72 m 1

m 1

Page 12

..

Relationship between price and expenditure gap • In a static utility maximization setting u  f , x   f  1   x  

F  pf 



1



 m   1    p 



1

p 



1 is the food-non food elasticity of substitution 1

F 

• The food price gap g  ln  H  and the expenditure gap G  ln  H   pL   FL  st • Approximation (1 -order) yields 

G

1    1 1      pL g   

– In an inter-temporal setting, the relationship will become non-linearly concave and heterogeneous depending on liquidity constraints, income levels and smoothing capacities (PIH will not apply) – Combination of liquidity constraints, income effects, and a static decreasing price elasticity with income may lead to an inverted U-shape pattern of expenditure gap with respect to income – Time-inconsistent preferences or habit formation? A structural framework can be estimated with the data we have : not the focus of this paper Page 13

Prima facie evidence on consumption seasonality (desc) 800

70%

700

60%

Tanzania P0

50%

Malawi P0

40%

Uganda P0

600 500 400

30% 300 200 100 0

20%

Tanzania 200809 Malawi 2010-11

10% Uganda 2009-10 0%

Page 14

Prima facie on seasonal volatility in agricultural markets Seasonal volatility shares in Tanzania 25%

20% 15% 10% 5% 0%

Cross-country for maize on wholesale markets: seasonality = around one quarter of overall price volatility in Tanzania and Uganda, but 40% in Malawi, where volatility is highest.

• In Tanzania, seasonality = 15% to 22% of the month-to-month total variability of prices, highest for maize, potatoes and sorghum, lower for other crops. • Other sources: inter-annual variability and irregular intraannual Maize: seasonality volatility share 50% 40% 30% 20% 10% 0% Malawi

Uganda

Tanzania Page 15

Market seasonality – specification issues • Non stationary I(1), AR(2) for most of them • Trend is stochastic in the majority of cases (constant drift most likely) but more for wholesale vs. retail markets • Seasonality in wholesale prices stat. Significant at the 5% level in most markets (marketplaces and crops), a bit less in retail prices – For almost all maize, rice, and beans markets – Much less often for cassava and starchy bananas (but still high levels)

• Seasonality is time-varying in a third of cases

Page 16

Findings – wholesale maize seasonal price gap between 30(TZ) and 50% (MWLI) on average; higher than rice, sorghum and millet Maize seasonality gap 60% 50% 40% 30%

20% 10% 0% Malawi

60%

Tanzania

50%

Uganda

Uganda

Tanzania

Malawi

40% 30% 20% 10% 0% Maize wholesale

Rice wholesale

Millet wholesale Sorghum wholesale Beans wholesale

Page 17

Findings- vegetable prices have highest gaps, cassava and groundnuts are of intermediate levels, wholesale > retail, and consistent ranking across countries 60% Tanzania

50%

Uganda

Malawi

40% 30% 20%

10% 0% Maize wholesale

Rice wholesale

Millet wholesale

Beans wholesale

Irish potatoes

Groundnuts

Cassava fresh

Tomatoes

60% Tanzania

50%

Uganda

Malawi

40% 30%

20% 10% 0% Maize wholesale

Maize retail Rice wholesale

Rice retail

Beans wholesale

Beans retail

Page 18

…but substantial within-country heterogeneity

0

.2

.4

.6

Wholesale maize seasonal conditional price gap (2000-2013) across markets

Malawi

Tanzania Unconditional gap

Uganda Conditional gap

Page 19

Decomposition of price conditional seasonal gaps (anova) 100% 90%

Location x market type x crop/product

80%

Market type x crop/product

70%

Location x crop/product

60%

Crop / product

50% 40%

Location x market type

30% Market type (retail/wholesale) 20% Marketplace and country (for the "all countries" series)

10%

0% All countries

Malawi

Tanzania

Uganda

Page 20

Seasonality price gaps are consistent with expectations Seasonality (conditional) gap premium wrt the average product 40% 30% 20%

10% 0% -10% -20% -30%

Page 21

SAFEX as a maize benchmark Maize price seasonality - Malawi & SAFEX 30% 20% 10% 0% -10% -20% -30%

Maize price seasonality - Tanzania & SAFEX 20% 15% 10% 5% 0% -5% -10% -15%

Safex

Domestic

Safex

Maize price seasonality - Uganda & SAFEX

Domestic

Dec

Nov

Oct

Sep

Aug

Jul

Jun

May

Apr

Mar

Feb

Jan

20% 15% 10% 5% 0% -5% -10% -15%

Safex

r = 0.876

Dec

Nov

Oct

Sep

Aug

Jul

Jun

May

Apr

Mar

Feb

Jan

r = 0.844

Domestic

• SAFEX in Johannesburg is the leading futures market for white maize. • Seasonality in Tanzania follows that on SAFEX with a two month lag but with twice the amplitude. • Seasonality in Uganda had a different temporal pattern. Page 22

Consumption conditional seasonality at national (rew.) level – base specification 160%

18%

140%

16%

Estimated gap (left scale)

14%

120%

12%

100%

10% 80%

Sample gap (left scale)

8% 60%

6%

40%

4%

20%

2%

0%

0% Food

Non food

Malawi 2010-11

Food

Non food

Uganda 2009-10

Food

Non food

Estimated gini (right scale)

Sample gini (right scale)

Tanzania 2008-09 Page 23

Consumption seasonality (estimated): urban>rural, non food > food, Malawi < TZ < UG 250% Estimated gap - urban

200%

150% Estimated gap - rural 1 harvest

100%

50%

Estimated gap - rural 2 harvest

0% Food

Non food

Malawi 2010-11

Food

Non food

Uganda 2009-10

Food

Non food

Tanzania 2008-09 Page 24

Robustness – specification of the RHS variables 120% 100%

Unconditional gap

80%

60%

Conditional gap with time-invariant controls only

40% 20% 0% Food

Non-food

Malawi

Food

Non-food

Tanzania

Food

Non-food

Conditional gap with income and asset controls included

Uganda

Page 25

Robustness – regional controls and sample weights • Correlation of seasonal factors at aggregated and sub-sample levels across the 4 specifications – Above 90% when urban and rural households considered separately (rural hh further separated out per regions with bimodal vs. unimodal rainfalls or by latitute-MLWI), between 20 and 80% at the aggregated level, above 70% at reaggregated level

• Joint significance of seasonal dummies always rejected for subsamples, not for the national and rural sample in Tanzania • Seasonality indicators very robust to specification when reweighted or reaggregated

Page 26

Estimated consumption cycles (smoothened) : Tanzania 2008-09 Evidence of heterogeneous timing and lack of smoothing 50% Urban hhds Non food

40% 30%

Rural bimodal Non food

20%

Rural unimodal Non food

10% 0% 1 -10% -20%

2

3

4

5

6

7

8

9

10

11

12

Urban hhds Food

Rural bimodal Food

-30% -40%

Rural unimodal Food

-50%

Page 27

Estimated consumption cycles (smoothened) : Malawi 2010-11 30% Urban Non food 20%

Rural North Non food Rural Central Non food

10%

Rural South Non food 0% 1

-10%

2

3

4

5

6

7

8

9

10

11

12

Urban Food Rural North Food Rural Central Food

-20% Rural South Food -30% Page 28

Estimated consumption cycles (smoothened) : Uganda 09-10 Evidence of intra-annual food-non food substitution 60% Urban hhds Non food

40% 20%

Rural bimodal Non food

0% 1 -20%

2

3

4

5

6

7

8

9

10

11

12 Rural unimodal Non food

-40% -60% -80%

-100% -120%

Urban hhds Food Rural bimodal Food Rural unimodal Food

-140%

Page 29

Internal validity and regular seasonality vs. intra-annual variability: the case of Tanzania 120%

100%

Estimated gap Tanzania 2008 survey

80%

60%

Estimated gap Tanzania 2010 survey

40%

20%

0% Urban sub sample

Rural with 2 harvests Food

Rural with 1 harvest

Urban sub sample

Rural with 2 harvests

Rural with 1 harvest

Estimated gap Tanzania pooled data

Non food Page 30

Robustness to RHS specification – TZ consumption cycles: urban (top)/national (bottom), food (right)/non food (left) 40%

Pooled Augmented without income

30%

Pooled Base conditional 2008-09 Augmented with income 2008-09 Augmented without income

0% 1

2

3

4

5

6

7

8

9

10

11

40% 30%

20% 10%

50%

20% 10% 0% 1

-10%

12

-10%

2008-09 Base conditional

-30%

-20%

2010-11 Augmented without income

-30%

2010-11 Base conditional

-50%

2

3

4

5

5

6

6

7

8

9

10

20%

15%

15%

10%

10%

-40%

5%

5%

0%

0% 1

2

3

4

5

6

7

8

9

10

11

12

1

2

3

4

7

8

9

10

-5%

-5% -10%

-15%

12

-20%

20%

-10%

11

-15% -20%

Page 31

11

12

Robustness to hh sub-sample categorization: rural sub-samples with and without cash crop growers: UG 2009-10 20% 10% 0% 1

2

3

4

5

6

7

8

9

10

11

12

-10% -20% Non food rural 2 harv Non food rural 2 non coffee Non food rural 2 no cash crop Food rural 2 harv Food rural 2 non coffee Food rural 2 no cash crop

-30% -40% -50%

25% 15% 5% -5%

1

2

3

4

5

6

7

8

9 10 11 12

Non food rural 1 harv Non food rural 1 non coffee Non food rural 1 no cash crop Food rural 1 harv

-15% -25% -35%

Food rural 1 non coffee Food rural 1 no cash crop

25% 20% 15% 10% 5% 0% -5% -10% -15% -20%

1

2 3 4 5 Rural cash crop non food Rural cash crop food

6

7

8

9

10

Page 32

11

12

Consumption seasonality tracks staple price seasonality; Higher contribution of food - TZ 15%

8% 6%

10%

4% 5%

2%

0%

0% 1

2

3

4

5

6

7

8

9

10

11

12 -2%

-5%

-4% -10%

-6%

-15% % change monthly deviation from annual consumption average

-8% Contribution of food consumption to total consumption change Contribution of non-food consumption to total consumption change % change monthly Change in total expenditures deviation from staple price average Month price effects (staple index) Page 33

Consumption seasonality tracks staple price seasonality; Less sensitivity to prices- MLWI 6%

20% 15%

4%

10% 2%

5%

0%

0% 1

2

3

4

5

6

7

8

9

10

11

12 -5%

-2%

-10% -4%

-6% % change monthly deviation from annual consumption average

-15% -20%

Contribution of food consumption to total consumption change Contribution of non-food consumption on total consumption change Change in total expenditures Price month effects (staple index)

% change monthly deviation from annual staple price average

Page 34

Food / non-food intra-annual substitution but highest non food consumption sensitivity to prices and contribution- UG 20%

8%

15%

6%

10%

4%

5%

2%

0%

0% 1

2

3

4

5

6

7

8

9

10

11

12

-5%

-2%

-10%

-4%

-15%

-6%

-20%

-8%

-25% % change monthly deviation from annual cons. average

Contribution of food consumption to total consumption change Contribution of non-food consumption to total consumption change Change in total expenditures Price month effects (staple index)

-10% % change monthly deviation from annual price average

Page 35

Matching with crop calendar data– MLWI

Page 36

Discussions and interpretation • Correlation in the production cycle of price and consumption patterns – Suggestive of strong irregular additional seasonality effects – Suggestive of strong income and level effects – Suggestive of some intra-annual food-non food substitution (at least in Uganda) – Strong heterogeneity across households and regions – Consistent with crop calendar data and differences between food and cash crop growers

• Inferences about seasonality in consumption difficult to draw from cross-sectional data but – Cross-country evidence show an evidence lack of most households’ consumption smoothing – Pooled data in TZ shows that there are some year-specific irregular and/or excessive seasonality e.g. 2008-09 vs. 2010-11 – May be worth exploring further the case of policy interventions (eg food security in Malawi) – Income effect shows an inverted U-shape with quantiles, suggestive of a Kuznets curve of seasonal inequality under high storage costs, lack of coping/saving and liquidity constraints – Habit formation to seasonal price spikes and seasonal demand shifts? Page 37

Conclusion • Seasonality still permeates African agriculture and livelihoods (also in people’s food consumption) : • On markets – Significant price gaps - 25% for maize -, consistent across countries + higher seasonality for wholesale vs. retail markets, especially for staples – Well correlated with the crop calendars / production cycles – Heterogeneity in price gaps explained chiefly by the type of crop/ product, and then in combination with locations (but very small pure location effects)

• To the households: – High sensitivity of consumption to the month dummies, rural
• Implications – Timing of consumption data collection still key for poverty calculations – Important agenda to better understand reasons for food price seasonality – Not only inter annual, but also intra-annual food consumption smoothing deserves more analytical and policy attention Page 38

Going further: applications of the new framework • Beyond the juxtaposition / correlation exercise – Once household panel data become available

• In-depth analysis of cyclical decomposition of intra-annual variability and regular vs. irregular seasonality (stochastic amplitude and phases) • Better understand / identify the general-equilibrium pass-through effects of market price seasonality across markets and households, • Measurement and identification of welfare and income effects – Child nutrition, Consumer welfare, and household behavior (structural approach) – Storage, purchasing, and marketing behavior -endogenous – Heterogeneity and quantile analysis (net seller/ net buyer, transient poverty)

• Identify the drivers of seasonality and mapping markets to geographic variables and population characteristics – Regular vs. exceptional seasonality (weather shocks, socio-economic shocks) – Policy interventions (food security and trade policies, ag. development, other safety nets or CCT programs…) – Market integration, credit and liquidity constraints – Crop calendar regular vs. year-specific data Page 39