1
New Home Sales across U.S. Regions and the Little Help from Monetary Policy.
André Varella Mollick*
Abstract: This paper examines the dynamics of new home sales across the four major U.S. regions - Midwest, Northeast, South, and West - using quarterly data from 1983:01 to 2006:01. While local income has grown stronger than average in the South and West, all areas should have benefited from the substantial fall in interest rates over the period of U.S. disinflation. Several results from conventional vector autoregression (VAR) models shed light on these two forces. First, for the whole U.S. new home sales respond strongly to real income and negatively to contractionary monetary policy shocks or to mortgage rates. Second, shocks to real income have positive and immediate impacts on regional new home sales. Third, negative and persistent impacts of FF shocks are found for the Midwest and West regions, with 30-year mortgage rate shocks causing negative and immediate effects in all but the Northeast. For the Midwest and West, falling interest rate shocks help new home sales increase. While in the Midwest the decline in rates alleviates the sluggish regional economy, in the West it amplifies the response of new home sales to a more vibrant economy. We discuss alternative channels of the transmission mechanism.
Keywords: monetary policy, new home sales, personal income, U.S. regions, VAR. JEL Classification Codes: E52, R31.
* Department of Economics and Finance, College of Business Administration, University of Texas-Pan American (UTPA), 1201 W. University Dr., Edinburg, TX 78539-2999, USA. E-mail:
[email protected] Tel.: +1-956-316-7913 and fax: +1-956-384-5020. I acknowledge able research assistance from Radhames Lizardo in the data construction.
2 1. Introduction The U.S. housing market exhibits extreme variations at the regional level, most likely because the intensity of economic activity vary from region to region at a given point in time. Several studies argue that the U.S. Midwest is more sensitive to monetary policy shocks that occur at the national level. The evidence on regional dynamics and monetary policy is perhaps better illustrated by the vector autoregressions (VAR) model by Carlino and DeFina (1998), which includes the growth rates of real personal income in the eight BEA regions for the U.S. and the relative price of energy. Their study identified one region (the Great Lakes) that is clearly more sensitive to monetary policy changes and two (Southwest and Rocky Mountains) that are much less sensitive. As Chiodo and Owyang (2003, p. 13) have observed, the larger effect on the Great Lakes region is explained by the fact that the region is “dependent on the manufacturing sector of its economy. The regions that are least sensitive to an increase in interest rates are the Southwest and Rocky Mountain regions, which have a more diverse combination of industries. The other five regions respond more closely to the U.S. average.” See also Owyang and Wall (2006). Press accounts report that the U.S. Midwest has a weaker economy and thus lower housing prices than other regions.1 Given the dynamic interrelations within a set of variables, the VAR model of the U.S. housing market in this paper complements existing single equation models. Examples of the latter group include the supply-determined model of the housing sector 1
The following recent quote from the WSJ (Sep 21, 2006) on the U.S. Midwest housing market is emblematic: “Home prices in the region have hardly budged over the past few years because of its weaker economy compared with other regions. Michigan, for example, has lost nearly 300,000 jobs since 2000, and its jobless rate has been consistently higher from the national average. A recent report by the OFHEO looked at housing prices in 275 metropolitan areas across the country. Six of the seven metropolitan areas that showed housing-price declines for the 12 months ended June 30 were in Indiana and Michigan. The study also stated that housing prices in states like Indiana, Ohio and Michigan were fairly flat over the past year but actually declined in the second quarter.”
3 by Topel and Rosen (1988), who report rapid adjustment speed and sizable long-run price-elasticity of supply for the quantity of new single-family housing units in the U.S., allowing for cost shifters such as interest rates. Gallin (2006) casts doubt on the assumption that housing prices and income are cointegrated for the U.S. McCarthy and Peach (2002) present a structural model as well with permanent income of households and the user cost of holding the housing asset in the demand side. Riddel (2004) offers an error correction modeling to housing dynamics in the U.S. and Iacoviello (2004) develops a general equilibrium model and provides support for housing prices driving consumption fluctuations in the U.S. VAR-related studies form another group of studies. Effects of monetary policy on the U.S. aggregate housing market are provided by McCarthy and Peach (2004) and Iacoviello (2005). VAR analysis on U.S. housing starts includes Ewing and Wang (2005) and, for the Singapore economy, Edelstein and Lum (2004) who consider four real endogenous variables (consumption, disposable income, private housing price and public housing wealth) and one exogenous variable (the level interest rate). The cross-country analysis in Tsatsaronis and Zhu (2004) show the dominance of inflation in the determination of real house prices in a SVAR model. Chirinko et al. (2004) explore the response of the economies of Europe, Japan, and the U.S. to shocks in housing and equity prices and find heterogenous impacts across countries. The model in this paper is methodologically based on Baffoe-Bonnie (1998) who shows that, like the result for the whole economy, a shock to the mortgage rate generates an immediate response in all regions. Several problems exist, however, with such
4 econometric methodology, as pointed out by Lastrapes (2002) in his structural VAR (SVAR) model for the aggregate U.S.2 The major features of the VAR in this paper, however, depart from Baffoe-Bonnie (1998) in at least three ways. First, we offer a blend of two literatures within the VAR methodology: i) the class of models in Christiano et al. (2005), among others, which take into account the transmission mechanism of monetary policy into output and prices; and ii) the VAR models in Carlino and DeFina (1998) and Clark (1998) of regional or industrial heterogeneity. Similar to Fratantoni and Schuh (2003), our recursive and structural VARs link national and regional markets by the mortgage rate, the channel of monetary transmission within the housing market. In contrast to their model, however, the dynamics herein is simpler, allowing for central bank policy reactions to income fluctuations. While the conventional Cholesky decomposition of shocks is employed in this paper, for robustness purposes we also investigate the generalized impulse response functions (GIRFs) by Pesaran and Shin (1998) and a structural model with short-run restrictions. Second, the basic identifying assumption herein is that the U.S. Federal Reserve responds to aggregate and local output fluctuations. Given the emphasis on monetary policy, we examine the period right after the end of the 1979-1982 when the U.S. Federal Reserve focused more closely on output and prices when setting the Federal Funds (FF)
2
Lastrapes (2002, p. 50) refers to the results in Baffoe-Bonnie (1998) as “ambiguous and difficult to interpret, since he pays very little attention to identifying the exogenous source of money shocks.”
5 target. In contrast to Baffoe-Bonnie (1998), the selection of the 1983-2006 assures a homogenous period in terms of U.S. monetary policy.3 Third, we adopt the four regions of the U.S. Census Bureau and allow regional dynamics to take place. Because supply and demand of housing depend heavily on idiosyncratic and regional factors, this study considers a multivariate model of housing which depends on regional income dynamics. We define an aggregate index of (real) personal income as the income of all other regions but the one investigated in a particular regional VAR (local income). Several features highlight the contrasting behavior of regional components and suggest a reconsideration of the role of monetary policy across regional housing markets. Table 1 contains a summarized chart of the key regional variables over the 1983:I to 2006:I period of analysis. The amount of new home sales has grown more than the national average of 91.49% for the Midwest (163.18%) and the West (105.78%). The Northeast has in fact decreased the amount of new home sales and the South has grown at about the same rate as the nation. Price movements have been wild as well, with the Northeast and West commanding higher price growth of more than 200% than the U.S. at 137.93%. Local personal income has grown faster than the national average of 81.31% in the South and West and slower than the average in the Midwest and Northeast. Finally, the aggregate of the other three states grows at a more orderly manner, which is expected for an aggregate measure of three regions. Figure 1 shows new home sales being affected by the 1990 recession and more recently at the start of the 2006-2007 housing market crisis. Overall, the South has grown 3
See Cook (1989) for an analysis of the “monetarist experiment” of 1979-1982, Goodfriend and King (2005) for varying degrees of credibility on monetary policy after the changes in 1979, and Owyang and Wall (2006) for breaking their sample along these lines.
6 more rapidly than other regions since the 1990 downturn. Figure 2 displays the price of new home sales starting at around the $69,200 - $78,800 in 1983 and reaching $334,600 in the Northeast or $330,000 in the West, against $247,700 for the whole country. Against this background, the period of analysis has also been subject of the remarkable disinflation of the U.S. economy. Interest rates have moved down in tandem, with U.S. Federal Funds (30-year mortgage rates) varying from 8.77% in 1983:I to 4.59% in 2006:I. Similar values for the came down from 13.03% in 1983:I to 6.24% in 2006:I. See Figure 3 for these co-movements. Given the preceding trends, the central message of this paper is that, if local income does not grow fast enough, the impact of national mortgage rates on new home sales has to be larger. This is particularly felt at the U.S. Midwest, in which the 163% increase in new home sales along with a sluggish local income (growing at 66% much less than the average of 81%) had to be helped by something else, such as monetary policy shocks. Using quarterly data from 1983:01 to 2006:01, several results from regional vector autoregression (VAR) models are reported. We find evidence in this paper in support of this hypothesis with the housing in the Midwest being substantially more affected by interest rate shocks than elsewhere. This does not mean, however, that the slowly growing region is the only one blessed by the fall in interest rates. The other case with amplified response of new home sales to mortgage rates was the U.S. West, which, however, had higher price and income growth. In other words, the more dynamic region also benefited greatly from the interest rate cuts. This article contains four more sections. Section 2 presents the data and Section 3 introduces the VAR models. Section 4 presents the results and Section 5 concludes.
7 2. The Data The data come from several data sources. The price of new home sales (PNHS) in quality-adjusted USD - as well as the amount of new home sales (NHS), in thousands - is taken
from
the
U.S.
Census
Bureau
(http://www.census.gov/const/www/newressalesindex.html).4 The base year for the indexes is 1996. The original frequency of the data is quarterly and the regional classification groups the states into 4 major regions according to the U.S. Census Bureau classification of “9 census regions”:5 Northeast comprises 9 states in the New England and Middle Atlantic regions: Connecticut, Maine, Massachussets, New Hampshire, Rhode Island, Vermont, New York, New Jersey, and Pennsylvania; Midwest comprises 12 states in the East North Central and West North Central regions: Ohio, Indiana, Illinois, Michigan, Wisconsin, Minnesota, Iowa, Missouri, North Dakota, South Dakota, Nebraska, and Kansas;
4
The Census Bureau collects NHS based upon the following definition: “A sale of the new house occurs with the signing of a sales contract or the acceptance of a deposit.” The house can be in any stage of construction: not yet started, under construction, or already completed (about 25% of the houses sold), which makes the previous two groups as evenly splitting the remaining 75% of houses. Existing home sales data are provided by the National Association of Realtors, in which “the majority of transactions are reported when the sales contract is closed.” Most transactions usually involve a mortgage which takes 3060 days to close. Therefore, an existing home sale (closing) most likely involves a sales contract that was signed a month or two. NHS usually lead existing home sales regarding changes in the residential sales market by a month or two: an existing home sale in January was probably signed 30 to 45 days earlier. Estimates of sales price of sold houses published by the Census Bureau are calculated from data acquired from interviews of home builders by the Survey of Construction. Two methods of median-estimation are considered: the first uses the sample weights to estimate medians via empirical cumulative-distribution functions and the second uses linear interpolation of grouped continuous data to approximate the median. See Thompson and Sigman (2001) for details. 5 This is similar to the regional classification undertaken by Del Negro and Otrok (2007) more recently. The (alternative) BEA classification has 8 regions: New England, Mideast, Great Lakes, Plains, Southeast, Southwest, Rocky Mountains, and Far West. Clark (1998) also employs the “9 census regions” classification with the following modifications from those listed above: D.C., Alaska, and Hawaii were excluded, the last two because employment figures were not available for 1947-1959.
8 South comprises 17 states in the South Atlantic, East South Central and West South Central regions: Delaware, Maryland, District of Columbia, Virginia, North Carolina, South Carolina, Georgia, Florida, West Virginia, Kentucky, Tennessee, Alabama, Mississippi, Arkansas, Louisiana, Oklahoma, and Texas; and West comprises 13 states in the Mountain and Pacific regions: Montana, Idaho, Wyoming, Colorado, New Mexico, Arizona, Utah, Nevada, Washington, Oregon, California, Alaska, and Hawaii. Personal income data come from Bureau of Economic Analysis (BEA) of the U.S. Dept. of Commerce (http://bea.gov/). The original figures were deflated by regional CPI indexes in order to obtain real personal income measures by the four Census regions. We bring the individual states from BEA into the region classification from U.S. Census Bureau outlined above, which ensures data consistency. For the aggregate VAR, the commodity prices variable (COM) is the PPI for all commodities (original code is PPIACO), not seasonally adjusted, 1982=100, originally in monthly values, from the Bureau of Labor Statistics (BLS) of the U.S. Dept. of Labor (http://www.bls.gov/). The price level variable (CPI), also not seasonally adjusted, 1982=100, originally in monthly values, also come from BLS. Both COM and CPI series were then aggregated into quarterly figures by averaging. The mortgage rate is from the Board of Governors of the Federal Reserve Board, available at the St. Louis FED (http://www.frbstlouis.com/) monthly, in % per annum, average of daily figures. The data are originally in monthly values and is for the 30-year conventional mortgage rate (original code is MORTG): average contract rate in commitments for fixed-rate first mortgages with permission from the Federal Home Loan
9 Mortgage. The federal funds rate (FF) is the effective federal funds rate (original code is FEDFUNDS), monthly, in % per annum, average of daily figures from the same sources. Both interest rate series were aggregated into quarterly figures by averaging. The original dataset was constructed for the 1973:I to 2006:I period. However, given the importance of monetary policy shocks to our strategy of contrasting regional to national factors, we focus on the 1983:I to 2006:I subsample. The reason for this is that it is very well established that after the “monetarist experiment of 1979” the U.S. Federal Reserve started to pay more attention to output and inflation in their decisions to change or hold FF rates. See Cook (1989) and Goodfriend and King (2005) for discussions of policymaking and McConnell and Perez-Quiros (2000) for formal evidence on a structural break in the volatility of U.S. GDP at around the breakpoint chosen herein. Table 2 contains the descriptive statistics for the U.S. and for the four Census regions. Home sales prices in Northeast (median of $339,000 in 2006) and West (median of $333,500) regions are much higher than Midwest (median of $197,700) and South (median of $195,400). As shown in a companion table to this paper available upon request, the correlation coefficient between aggregate U.S. new home sales and the 30year mortgage rate is -0.693 and between new home sales and personal income is +0.849. Negative correlation coefficients between new home sales and mortgage rates are seen at the regional level. We construct the index of aggregate personal income taking all other three regions and excluding region “i”. Local income is highly correlated with aggregate income. For example, personal income in the Northeast (YPNE) has a 0.988 correlation coefficient with personal income of the remaining three regions (YPUSNE). Similar
10 coefficients are found for the other regions. In terms of quarterly growth rates, mean personal income grew at 0.66% for the whole U.S., 0.55% for the Midwest, 0.50% for the Northeast, and faster by 0.80% for the South and 0.74% for the West. Figure 2 displays the rising price trend of the new home sales. In terms of quarterly growth rates, mean prices grew at 1.34% for the whole U.S., 1.92% for the Northeast, 1.62% for the West, against the more subdued 1.18% for the Midwest and 1.17% for the South. Similar trends are obtained from with OFHEO’s House Price Index (HPI), which is a repeat-sales index.6 The share of the South region in U.S. personal income is growing over time, reaching 0.351 in the first quarter of 2006, from 0.308 in the first quarter of 1983. The U.S. West region share also grows over the three decades, setting at 0.230 more recently. Both Northeast (from 0.243 to 0.223) and Midwest (from 0.231 to 0.201) lost importance over time.
3. The Empirical Models Following Lastrapes (2002) and Christiano et al. (2005), we identify U.S. monetary policy shocks with the disturbance term in the regression equation:
rt = F (Φt) + εrt
6
(1),
For comparison, we aggregated OFHEO’s data (available at http://www.ofheo.gov/HPI.asp) originally from the nine U.S. Census Bureau classifications into the four Census regions discussed above. We then compared our measure of prices on NHS to OFHEO’s HPI and got trends very similar to those in the upper part of Figure 1. The correlation coefficients between PNHS and HPI were very high across the four regions: 0.984 for the Midwest; 0.965 for the Northeast; 0.981 for the South; and 0.987 for the West. We will thus focus on the U.S. Census Bureau housing figures. For the differences between the OFHEO’ price and the constant quality house price index by the U.S. Census Bureau, see McCarthy and Peach (2004).
11 where: rt is the policy instrument or the Federal Funds target rate as in Bernanke and Blinder (1992), F is a linear function, Φt is the information set, and εrt is a serially uncorrelated shock orthogonal to the elements of Φt. For εrt to be an exogenous policy shock, (1) must be viewed as the authority’s decision rule for setting the instrument. The orthogonality conditions on εrt correspond to the assumption that date t policy shocks do not affect the elements of Φt. Of course, εrt could also reflect errors in the specification of the decision rule due to changing central bank’s rules over time or to nonlinearities. Similar to Christiano et al. (2005), we partition the vector of variables Yt as Yt = [Y1t, rt, Y2t], where Y1t contains the variables whose time t values are contained in Φt and that are assumed not to respond contemporaneously to a monetary policy shock (yet can contemporaneously affect such policy). The vector Y2t consists of the time t values of all the other variables in Φt; these variables do not have an impact on Y1t or rt. In our context, Y1t is composed of real national income aggregates excluding region “i” (YP-i), and real region’s “i” real income (YPi). Vector Y2t contains variables related to each local housing market, such as the price of new home sales (PNHSi) and the amount of new home sales (NHSi).7 The separation of personal income into regional (local) income and an aggregate of all other income follows from Carlino and DeFina (1998), who allow for output effects from region to region. The omission of aggregate income would cast doubt on the 7
With the exception of the regional income variables, a similar SVAR is estimated by Lastrapes (2002) as follows: Yt = [Y1t, rt, Y2t], with Y1t containing a commodity price index; (national) output, the price level, and rt standing for the federal funds rate; and Y2t containing total reserves in the banking system, nonborrowed reserves, the mortgage rate, real house prices, and the flow of house sales. Commodity prices (COM) were used in previous VARs of this paper but are not essential in our application to the housing market and were excluded for parsimonious reasons. Similarly, the price level (CPIi) was included in augmented models, although its presence is much more important for the national VAR since the FED is not going to focus on regional CPI on setting the FF rate. The inclusion of CPIi in regional VARs does not change the results and are not statistically significant in general.
12 identification hypothesis since the FED has to monitor aggregate income growth when setting interest rates. For this reason, we combine local and aggregate forces in the VAR model. In our specification local income forces are contemporaneously independent of policy variables (interest rates) and of (regional) housing market variables. The FED, however, observes national and local income dynamics when setting short-term interest rates. The benchmark model in levels is Y = [YP-i, YPi, FF, R30, PNHSi, NHSi]’, where ’ represents the transpose sign. Assume that the vector (n x 1) Yt of endogenous variables is generated by the following structural model:
A0Yt = A1Yt-1 + … + ApYt-p + ut
(1),
where ut is an (n x 1) vector of serially and contemporaneously uncorrelated shocks with unit variance. The implied moving average representation is:
Yt = (D0 + D1L + D2L2 + …) ut
or
Yt = D(L)ut
(2),
where D(L) = (A0 - A1L - …- ApLp)-1 and L denotes the lag operator. The moving average of the model’s reduced form is:
Yt = (I + C1L + C2L2 + …) εt
or
Yt = C(L)εt
(3),
13
where εt = D0ut, Ci = DiD0-1, and E εtεt’ ≡ Σ = D0D0’. While the reduced form parameters C(L) and Σ are estimable from the VAR, E εtεt’ ≡ Σ = D0D0’ is not a unique mapping from structural to the reduced form. The identification strategy imposes a sufficient number of restrictions on D0 to identify the structural coefficients from C(L) and Σ. In the benchmark case, the restrictions make D0 lower block recursive, such that the zero restrictions just-identify responses to monetary policy shocks. Impulse responses and variance decompositions can then be computed and inference be drawn. In recursive VARs, the Cholesky decomposition of shocks is employed in order to orthogonalize the errors of each equation as surveyed in Stock and Watson (2001). For robustness, we implement two additional methodologies: the generalized impulse response functions (GIRFs) by Pesaran and Shin (1998) and a structural model with short-run restrictions. Both did not change qualitatively the basic pattern below based on the Cholesky decompositions.8 Associated with each equation in the VAR are the shocks, which can be given an intuitive interpretation as follows. Technological improvements to aggregate or local 8
In addition to the GIRFs, we also employ the SVAR alternative with 6 short-run restrictions. The SVAR will vary from region to region depending on the data-supported restrictions. Let A and B be the (k x k) matrices, ε be the vector of observed (reduced-form) residuals and u be the vector of unobserved (structural) innovations. Let Aεt = But, where the structural innovations are orthonormal: E utut’ = I. Restrictions which just-identify the SVAR are as follows: ε1 = C(1)u1 (income of aggregate regions) ε2 = C(2)ε1 + C(3)u2 (income of region i) ε3 = C(4)ε1 + C(5)ε2 + c(6)u3 (FF rate) ε4 = C(7)ε1 + C(8)ε2 + c(9)ε3 + c(10)u4 (mortgage rate) ε5 = C(11)ε1 + C(12)ε2 + c(13)ε3 + c(14)ε4 + c(15)u5 (price of new home sales of region i) ε6 = C(16)ε1 + C(17)ε2 + c(18)ε3 + c(19)ε4 + c(20)ε5 + c(21)u6 (new home sales of region i) The SVAR model imposes some of the C (.)’s coefficients equal to zero, depending on what is estimated for each region. For example, in the SVAR for the U.S. Midwest the last 2 rows become ε5 = c(14)ε4 + c(15)u5 and ε6 = C(16)ε1 + c(19)ε4 + c(21)u6. We then test for the overidentified restrictions under the null hypothesis that the restrictions are valid.
14 income (ε1 and ε2 for the illustrative model in the previous footnote) are the main force underlying output changes as in real business cycle (RBC) models. We estimate with aggregate personal income preceding local personal income, but the estimations are robust to changes in this particular ordering. The model claims that information on real income is taken into account by the Fed when setting interest rates apart from interest rate smoothing. This follows the growing literature on “Taylor rules” in Orphanides (2001) or on the international transmission of monetary shocks in Kim (2001). Interest rate shocks include permanent disturbances to central bank policymaking due to human actions or to shifts in general economic efficiency, such as the more aggressive FED policymaking in the Volcker-Greenspan years.9 We will refer to ε3 as monetary policy shocks. Disturbances to the mortgage interest rate (ε4) can be viewed as shifts in the housing market of funds. Like the federal funds rate, shocks to mortgage rates respond to output fluctuations. Housing market variables have their corresponding shocks given by ε5 and ε6, of which the latter is assumed to respond to all forces. Although Fed officials are increasingly monitoring the housing market, we do not use this assumption in the identifying assumptions.10
9
Clarida et al. (2000) contains evidence that the U.S. Federal Reserve has become stricter in fighting inflation since 1979, when nominal rates started to move up more than proportionately with inflation. 10
It is becoming more common for Fed officials to allude to the housing market, such as Fed Vice Chairman Donald Kohn on January 8, 2007, who stroke a note of caution: “Tentative signs have begun to emerge that the housing market may be stabilizing … Housing starts may be not very far from their through … But the risks around this outlook still are largely to the downside.” (The Wall Street Journal, Jan 9, 2007). A contrary view was provided by Federal Reserve Governor Frederic Mishkin in prepared remarks to the Forecasters Club of New York on January 17, 2007: "Central banks should not give a special role to house prices in the conduct of monetary policy but should respond to them only to the extent that they have foreseeable effects on inflation and employment." Citing examples in Europe when policy makers seemed to use house prices as a justification for an interest-rate decision, he argued that central banks have no better means of identifying potential bubbles than private markets do, according to the WSJ on January 17, 2007.
15
4. Results Preliminary tests do reject the null of no cointegration vectors for the Trace test, but the results are reversed under the Maximum Eigenvalue test. Since the evidence of cointegration is not strong, we proceed with the impulse responses for the VAR in levels, which conforms to the tradition of most of the literature of monetary policy shocks in Lastrapes (2002) or Christiano et al. (2005). In general, the series were found to be nonstationary in levels and stationary in first-differences. Logarithms were taken on all variables except for the interest rates, following common practice. We employ the benchmark VAR model explained in Section 3 with three seasonal dummies. The VAR is estimated initially with 4 lags and information criteria are used to decide the final lag length, together with Lagrange multiplier (LM) tests on the residuals. After experimentation, we settled on either k = 2 or k = 3 lag-lengths. We run VAR models for the U.S. as a whole first. In all cases, given the seasonality of housing markets, we include three quarterly seasonal dummies in the VAR. Since commodity prices do not solve the “price puzzle” (the U.S. CPI responds positively to FF rate shocks for until about 6 quarters), we omit commodity prices in the subsequent analysis. In fact, in the context of this research nothing is gained when we enlarge the benchmark 6-variable VAR to a 7-variable VAR with commodity prices. We focus on the responses of new home sales. Since there is not much insight from shocks to prices, the responses to shocks in the price of new home sales are not reported. Also, given that the responses to new home sales to their own shocks decay
16 very fast from very high levels at the first quarter only, we omit these impulse responses. We therefore henceforth refer to the remaining impulse responses. Figure 4 shows the impulse responses of new U.S. home sales to income shocks (YP), monetary policy shocks (FF), and mortgage rate shocks (R30). Interestingly, we find the expected impulse responses for the U.S. as a whole. In response to a one-percent standard deviation in output shocks, there is a short-term positive effect on new home sales of 0.02 (standard deviation of 0.006); federal funds rate shocks have negative effects on new home sales varying from -0.018 (standard deviation of 0.007) at the fourth quarter to -0.021 (standard deviation of 0.011) at the ninth quarter; while mortgage shocks have strong negative effects on new home sales on impact, peaking at -0.032 (standard deviation of 0.007) at the second quarter. The variance decomposition of new home sales at the U.S. aggregate in Table 3 show the importance of these factors throughout, with real income ending the 3-year forecasted horizon explaining about 24% of the variance in new home sales. Similar figures for the FF and R30 rates are 21% and 15%. These values suggest strong long-run effects on the national VAR of all shocks. A detailed check of Figure 4 suggests that income shocks are positive and significant throughout the estimation period. With the increase in income, the amount of new home sales transacted increases at all forecasted horizons. Besides, the housing response is more than proportional to the shock to one standard deviation in income. After a one year lag, contractionary monetary policy shocks have a negative effect on the housing market. The negative effects remain well within the second year. This implies that contractionary monetary policy shocks take a fairly long time to exert a negative impact on the price of new home sales. Similarly, mortgage shocks have immediate
17 negative effects at -0.018 (standard error of 0.006), notably at -0.032 (standard deviation of 0.007) at the second quarter, and at -0.018 (standard deviation of 0.008) at the third quarter. Figures 5 to 8 show the impulse responses of new home sales in each U.S. region to each of the four major shocks. We now have disaggregated aggregate and local income, which provide information to FED authorities when setting short-term rates at the national level. As expected, the response of housing markets to monetary policy forces vary from region to region. As the top left chart of Figure 5 shows for the U.S. Midwest, a one standard deviation to aggregate output leads to a very significant 0.022 (standard deviation of 0.009) response in new home sales at the first quarter, which remains statistically significant again at quarter 3. Local income is also marginally significant with the estimated impact of 0.019 (standard deviation of 0.011) at the third quarter on new home sales. This suggests that aggregate and, to a less extent, local income play an important role in explaining the variance of housing figures at the U.S. Midwest. Carlino and DeFina (1998) and Owyang and Wall (2006) have shown that income in this region is more responsive to monetary policy shocks and press accounts have speculated on local housing market responses. Using VARs with sign restrictions, Vargas-Silva (2008) documents that only in the Northeast and Midwest the response of housing starts to a contractionary monetary policy is negative for the 1965 – 2005 period with monthly (and interpolated) observations. Similar to the aggregate case, the negative effects of a contractionary monetary policy shock (a rise in FF shock) are visible after some time, with the negative impacts of -0.018, with standard deviations of 0.009 in quarter 6 and 0.010 in quarter 7. Similar to
18 the national VAR, the impact of shocks to the mortgage market is negative at the second quarter only: -0.032 (standard deviation of 0.010). The magnitude is, however, very strong, suggesting that new home sales are increasing in the Midwest in response to the declines in interest rates. The point response for a 1% increase in contractionary mortgage shocks is a fall of -3.2% in the U.S. Midwest. Therefore, for the U.S. Midwest, income shocks have positive effects on impact and monetary policy shocks have a negative effect on new home sales at the mid-horizon. The variance decompositions in Table 3 show that aggregate income shocks for the Midwest are responsible for 17.11% of new home sales after 3 years of the shock, about the same from 17.24% at the 6-quarter forecasted horizon. Local income shocks explain about 16.71% of the variance in the same horizon. Monetary policy shocks, however, explain only 15.35% of the variance decomposition of new home sales after 3 years of the shock and mortgage shocks explain 9.12% after 3 years of the shock. The combination of FF and R30 shocks helps explain about one fourth of the variance decomposition for the Midwest region. Figure 6 shows the impulse responses of new home sales in the U.S. Northeast to the same forces. Other than local income on impact, there are relatively muted responses for the Northeast region. The major result is that a one standard deviation to aggregate output leads to a significant 0.033 (standard deviation of 0.014) response in new home sales at the first quarter, which then fades to a zero response in the second quarter. The variance decompositions in Table 3 show that aggregate income shocks are responsible for only 4.02% of new home sales after 3 years of the shock, fairly stable across
19 forecasted horizons. For the U.S. Northeast region, new home sales is still very much (78%) explained by its own fluctuations in the long run. The responses in Figure 7 are a lot different for the South region since aggregate income shocks have a slightly positive impact of 0.013 (standard deviation of 0.006) on new home sales at the first quarter and 0.016 (standard deviation of 0.008) at the second quarter. Local income shocks have a delayed impact, however, after the fifth quarter: there are the positive impacts of 0.021 or 0.023 (with standard deviations of 0.010 or 0.012) lasting until quarter 9. There are no statistically significant monetary policy shocks for new home sales of the South region. One can note, however, that the impact of shocks to the mortgage market is negative and strong on impact and for the second quarter at 0.022 (standard deviation of 0.008). As far as the variance decompositions in Table 3 for the South, local income explains about 21.95% of the variance of home sales in the longrun. Mortgage rate shocks contribute to 12.23% of the variance in new home sales in U.S. South after 12 quarters of the shock, complementing the role of local income. For the West region in Figure 8 a one standard deviation shock to aggregate output leads to a 0.022 (standard deviation of 0.009) response in new home sales at the first quarter. The response to aggregate output shocks is prolonged, lasting until quarters 4 to 6. The new home sales response to local income shocks is also significant and persistent at 0.025 (standard deviation of 0.013) in the second quarter. Contractionary monetary policy shocks have a significant and long-lasting negative impact until quarter 9, peaking at quarters 7 to 9 with a -0.030 response (standard deviations of about 0.013 or 0.014). The impact of shocks to the mortgage market is negative at quarters 1 at -0.030 (standard deviation of 0.009) and 2 at -0.042 (standard deviation of 0.009). As in the
20 Midwest, the point response to a 1% increase in contractionary mortgage shocks is a fall of -4.2% in new home sales of the U.S. West. In addition to the robust aggregate income effects, new home sales in the West gets help from monetary policy shocks. The variance decompositions show that aggregate income shocks are responsible for slightly over 14% of new home sales in the West after 3 years of the shock. Monetary policy and mortgage rate shocks are very much important as well at the three-year forecasted horizon with over one quarter of the variance of new home sales in West U.S. for FF shocks and 12.53% for mortgage shocks. Our results indicate that the U.S. Midwest is the most strongly affected by output fluctuations, followed by the U.S. West, while contractionary monetary policy exert negative impacts at new home sales at the U.S. Midwest and West. The income and monetary policy shocks result on the U.S. Midwest has plenty of support in work by Carlino and DeFina (1998) and Owyang and Wall (2006). None of them has dealt with the housing market, however. Yet the results on the West may be more difficult to rationalize within the “industry mix” framework. We explore a couple of mechanisms. First, the effects of income on housing prices may be informative as in Iacoviello (2005).11 These could be found in our regional approach only for the Midwest with shocks in aggregate income explaining 27.78% of housing price fluctuations, against 8.16% for the whole U.S.
11
The study of monetary policy on the U.S. aggregate housing market by Iacoviello (2005) has shown that there is a two-way interaction between housing prices and output. These were also found in our regional approach with a stronger effect from output to the price of new home sales than from price of new home sales to output. The variance decompositions of the price of new home sales (not reported) confirm the very important role of income shocks in the variance of the price series in the Midwest at 27.78%, against 15.20% in the Northeast, 11.81% in the South, and only 5.80% in the West.
21 Second, Figure 9 has the combined impulse responses of home prices to shocks in new home sales for the two cases in which they happened to be statistically significant. In contrast to the lack of responses in the first three regions, the West shows very strong and positive responses on prices due to higher new home sales. The upper chart of Figure 9 has the response for the whole of the U.S, which is positive between quarters 4 and 8. The impulse response for the West at the bottom of Figure 9 is the only regional one that resembles this. Being persistent over almost the whole of the forecasted horizon, it is statistically significant starting at quarter 5 at +0.011 (standard deviation of 0.004) and goes on as well within the three year forecasted horizon. The variance decompositions (not reported) confirm the very important role of price shocks in the variance of new home sales in the West at 23.84%, against 0.47% in the Midwest, 10.08% in the Northeast, and 2.10% in the South. The similar national figure was 12.08%. Overall, this positive response of pricing to higher new home sales is consistent with a “bubble-like” pattern of prices in the West not falling after an increase in transacted homes.
5. Concluding Remarks The VAR models examine new home sales across the four U.S. regions using quarterly data from 1983:01 to 2006:01 with standard identification procedures and minimal identifying assumptions. Recent evidence in Gallin (2006) casts doubt on the assumption that housing prices and income are cointegrated for the U.S. The VAR models herein have sufficiently enough dynamic interactions within the variables. Allowing for aggregate and local real personal income similar to Carlino and DeFina (1998) and Owyang and Wall (2006), our approach deviates from Baffoe-Bonnie (1998)
22 and Vargas-Silva (2008) and examines the period right after the end of the 1979-1982 period when the U.S. Federal Reserve finalized its monetary targeting. With two recessions in 1991 and 2001, our sample period is characterized by a sustained decline in interest rates within uniform monetary policy procedures. Using quarterly data from 1983:01 to 2006:01, the vector autoregression (VAR) models gauge the forces of regional income against interest rates at the regional level. While income has grown stronger than average in some regions, interest rates have fallen substantially over the period following the U.S. disinflation period. We find negative and persistent impacts of FF shocks for the Midwest and West regions, with 30-year mortgage rate shocks having negative and immediate effects in all but the Northeast. The Midwest response is consistent with the less diversified “industry mix” advanced by Carlino and DeFina (1998), amplifying the transmission channel of monetary policy. For the Midwest, falling interest rates have a counteracting force in helping new home sales to increase given sluggish income movements. The U.S. West housing market, however, with higher price and income growth over the period, also benefited greatly from the interest rate cuts. It is, in fact, consistent with a “bubble-like” pattern of prices in the West not falling after an increase in the number of transacted homes. As an extension, Fed officials are increasingly monitoring the housing market, which has been suffering a strong correction lately.12 The controversial issue of the Federal Reserve responding to the housing market is left for future research.
12
The latest numbers are in line with the results of this paper with the West defying the downward trend. On the January 2008 data, “sales of new single-family homes decreased 2.8% last month to a seasonally adjusted annual rate of 588,000, the Commerce Department said. Sales tumbled 4% in December and 13.1% in November. Regionally, new home sales decreased 2.4% in the South, 7.6% in the Midwest and 10% in the Northeast. Sales rose 2.2% in the West.” (WSJ, February 28, 2008, p. A2).
23 References Baffoe-Bonnie, John, 1998, The Dynamic Impact of Macroeconomic Aggregates on Housing Prices and Stock of Houses: A National and Regional Analysis, Journal of Real Estate Finance and Economics 17 (2), 179-197. Bernanke Ben, and Alan Blinder, 1992, The Federal Funds Rate and the Channels of Monetary Transmission. American Economic Review 82 (4): 901-92. Carlino, Gerald, and Robert DeFina, 1998, The Differential Regional Effects of Monetary Policy. Review of Economics and Statistics 80 (4): 572 -587. Chiodo, Abbigail, and Michael Owyang, 2003, Monetary Policy: The Whole Country Gets the Same Treatment but Results Vary. The Regional Economist, January: 12 -13. Chirinko, Robert, Leo de Haan, and Elmer Sterken, 2004, Asset Price Shocks, Real Expenditures, and Financial Structure: A Multi-Country Analysis. Emory University, Working Paper. Christiano, Lawrence, Martin Eichenbaum, and Charles Evans, 2005, Nominal Rigidities and the Dynamic Effects of a Shock to Monetary Policy. Journal of Political Economy 113 (1): 1-45. Clarida Richard, Jodi Galí, and Marc Gertler, 2000, Monetary Policy Rules and Macroeconomic Stability: Evidence and Some Theory. Quarterly Journal of Economics 115 (1): 147-180. Clark, Todd, 1998, Employment Fluctuations in U.S. Regions and Industries: The Roles of National, Region-Specific, and Industry-Specific Shocks. Journal of Labor Economics 16: 202-229. Cook, Timothy, 1989, Determinants of the Federal Funds Rate: 1979-1982, Federal Reserve Bank of Richmond Economic Review, January/February, 3-19. Available at: http://www.richmondfed.org/publications/economic_research/economic_review/pdfs/er7 50101.pdf Del Negro, Marco, and Christopher Otrok, 2007, 99 Luftballoons: Monetary Policy and the
24 House Price Boom across U.S. States, Journal of Monetary Economics 54 (7), 1962 -1985. Edelstein, Robert, and Lum, Sau Kim, 2004, House Prices, Wealth Effects, and the Singapore Macroeconomy, Journal of Housing Economics 13, 342-367. Ewing, Bradley, and Yongsheng Wang, 2005, Single Housing Starts and Macroeconomic Activity: An Application of Generalized Impulse Response Analysis, Applied Economics Letters 12, 187-190. Fratantoni, Michael, and Scott Schuh, 2003, Monetary Policy, Housing, and Heterogenous Housing Markets, Journal of Money, Credit, and Banking 35 (4), 557-589. Gallin, Joshua, 2006, The Long-Run Relationship between House Prices and Income: Evidence from Local Housing Markets. Real Estate Economics 34 (3), 417-438. Goodfriend, Marvin, and Robert King, 2005, The Incredible Volcker Disinflation, Journal of Monetary Economics 52:981-1015. Iacoviello, Matteo, 2005, House Prices, Borrowing Constraints, and Monetary Policy in the Business Cycle, American Economic Review 95 (3), 739-764. Iacoviello, Matteo, 2004, Consumption, House Prices, and Collateral Constraints: A Structural Econometric Analysis, Journal of Housing Economics 13, 304-320. Kim, Soyoung, 2001, International Transmission of U.S. Monetary Policy Shocks: Evidence from VAR’s. Journal of Monetary Economics 48: 339-372. Lastrapes, William, 2002, The Real Price of Housing and Money Supply Shocks: Time Series Evidence and Theoretical Simulations, Journal of Housing Economics 11, 40-74. McCarthy, Jonathan, and Richard Peach, 2004, Are Home Prices the Next “Bubble”?, Federal Reserve Bank of New York Economic Policy Review, December, 1-17. McCarthy, Jonathan, and Richard Peach, 2002, Monetary Policy Transmission to Residential Investment, Federal Reserve Bank of New York Economic Policy Review 8 (1), December, 139-158.
25 McConnell, Margaret, and Gabriel Pérez-Quiros, 2000, Output Fluctuations in the United States: What has Changed since the Early 1980’s? American Economic Review 90 (5): 14641476. Orphanides Athanasios, 2001, Monetary Policy Rules Based on Real Time Data. American Economic Review 91 (4): 964-985. Owyang, Michael, and Howard Wall, 2006, Regional VARs and the Channels of Monetary Policy, Federal Reserve Bank of St. Louis Working Paper, 2006-002A. Available at: http://research.stlouisfed.org/wp/2006/2006-002.pdf Pesaran, H, and Y Shin, 1998, Generalized Impulse Response Analysis in Linear Multivariate Models. Economics Letters 58, 17-29. Riddel, Mary, 2004, Housing Market Disequilibrium: An Examination of Housing-Market Price and Stock Dynamics: 1967-1998, Journal of Housing Economics 13, 120-135. Stock, James, and Marc Watson, 2001, Vector Autoregressions. Journal of Economic Perspectives 15 (4): 101-115. Thompson, Katherine, and Richard Sigman, 2001, Estimation and Replicate Variance Estimation of Median Sales Prices of Sold Houses, U.S. Census Bureau. Available online at: http://www.census.gov/const/med_sumt.pdf Topel, Robert, and Sherwin Rosen, 1988, Housing Investment in the United States, Journal of Political Economy 96 (4), 718-740. Tsatsaronis, Kostas, and Haibin Zhu, 2004, What Drives Housing Price Dynamics: CrossCountry Evidence, BIS Quarterly Review, March, 65-78. Vargas-Silva, Carlos, 2008, Monetary Policy and the U.S. Housing Market: A VAR Analysis Imposing Sign Restrictions, Journal of Macroeconomics, forthcoming. The Wall Streeet Journal, various issues.
26
Figure 1. New Home Sales (PNHS) in thousands of units across U.S. Regions: 1983:I to 2006:I.
700 600 500 400 300 200 100 0 84
86
88
90
92
94
NHSMW NHSS
96
98
00
NHSNE NHSW
02
04
27 Figure 2. Price of New Home Sales (PNHS) in Constant-Quality USD across U.S. Regions: 1983:I to 2006:I.
400000 350000 300000 250000 200000 150000 100000 50000
84
86
88
90
92
94
PNHSMW PNHSNE
96
98
00
02
PNHSS PNHSW
04
28 Figure 3. The Federal Funds Rate (FF) and the 30-Year Mortgage Rate (R30) in the U.S.: 1983:I to 2006:I.
16 14 12 10 8 6 4 2 0 84
86
88
90
92
94 FF
96 R30
98
00
02
04
29
Figure 4. Impulse Responses of New Home Sales in the U.S. to Shocks in: Personal Income (YPUS), FF Rate (FF), and to 30-Year Mortgage Rates (R30).
Response to Cholesky One S.D. Innovations ± 2 S.E. Response of LOG(NHS) to LOG(YPUS)
Response of LOG(NHS) to FF
.06
.06
.04
.04
.02
.02
.00
.00
-.02
-.02
-.04
-.04
-.06
-.06 2
4
6
8
10
12
2
4
6
Response of LOG(NHS) to R30 .06 .04 .02 .00 -.02 -.04 -.06 2
4
6
8
10
12
8
10
12
30 Figure 5. Impulse Responses of New Home Sales in U.S. Midwest (NHSMW) to Shocks in: Aggregate Personal Income (YPUSMW), Local Personal Income (YPMW), FF Rate (FF), and to 30-Year Mortgage Rates (R30).
Response to Cholesky One S.D. Innovations ± 2 S.E. Response of LOG(NHSMW ) to LOG(YPUSMW )
Response of LOG(NHSMW ) to LOG(YPMW )
.12
.12
.08
.08
.04
.04
.00
.00
-.04
-.04
-.08
-.08 1
2
3
4
5
6
7
8
9
10 11 12
1
Response of LOG(NHSMW) to FF
3
4
5
6
7
8
9
10 11 12
Response of LOG(NHSMW ) to R30
.12
.12
.08
.08
.04
.04
.00
.00
-.04
-.04
-.08
2
-.08 1
2
3
4
5
6
7
8
9
10 11 12
1
2
3
4
5
6
7
8
9
10 11 12
31 Figure 6. Impulse Responses of New Home Sales in U.S. Northeast (NHSNE) to Shocks in: Aggregate Personal Income (YPUSNE), Local Personal Income (YPNE), FF Rate (FF), and to 30-Year Mortgage Rates (R30).
Response to Cholesky One S.D. Innovations ± 2 S.E. Response of LOG(NHSNE) to LOG(YPUSNE)
Response of LOG(NHSNE) to LOG(YPNE)
.15
.15
.10
.10
.05
.05
.00
.00
-.05
-.05
-.10
-.10 1
2
3
4
5
6
7
8
9
10 11 12
1
Response of LOG(NHSNE) to FF
3
4
5
6
7
8
9
10 11 12
Response of LOG(NHSNE) to R30
.15
.15
.10
.10
.05
.05
.00
.00
-.05
-.05
-.10
2
-.10 1
2
3
4
5
6
7
8
9
10 11 12
1
2
3
4
5
6
7
8
9
10 11 12
32 Figure 7. Impulse Responses of New Home Sales in U.S. South (NHSS) to Shocks in: Aggregate Personal Income (YPUSS), Local Personal Income (YPS), FF Rate (FF), and to 30-Year Mortgage Rates (R30).
Response to Cholesky One S.D. Innovations ± 2 S.E. Response of LOG(NHSS) to LOG(YPUSS)
Response of LOG(NHSS) to LOG(YPS)
.08
.08
.06
.06
.04
.04
.02
.02
.00
.00
-.02
-.02
-.04
-.04 1
2
3
4
5
6
7
8
9
10 11 12
1
2
Response of LOG(NHSS) to FF
4
5
6
7
8
9
10 11 12
Response of LOG(NHSS) to R30
.08
.08
.06
.06
.04
.04
.02
.02
.00
.00
-.02
-.02
-.04
3
-.04 1
2
3
4
5
6
7
8
9
10 11 12
1
2
3
4
5
6
7
8
9
10 11 12
33 Figure 8. Impulse Responses of New Home Sales in U.S. West (NHSW) to Shocks in: Aggregate Personal Income (YPUSW), Local Personal Income (YPW), FF Rate (FF), and to 30-Year Mortgage Rates (R30).
Response to Cholesky One S.D. Innovations ± 2 S.E. Response of LOG(NHSW) to LOG(YPUSW)
Response of LOG(NHSW) to LOG(YPW)
.12
.12
.08
.08
.04
.04
.00
.00
-.04
-.04
-.08
-.08 1
2
3
4
5
6
7
8
9
10 11 12
1
Response of LOG(NHSW) to FF
3
4
5
6
7
8
9
10 11 12
Response of LOG(NHSW) to R30
.12
.12
.08
.08
.04
.04
.00
.00
-.04
-.04
-.08
2
-.08 1
2
3
4
5
6
7
8
9
10 11 12
1
2
3
4
5
6
7
8
9
10 11 12
34 Figure 9. Impulse Responses of Prices of New Home Sales (PNHS) to Shocks in New Home Sales (NHS) in the whole U.S. and in the U.S. West (NHSW).
Response to Cholesky One S.D. Innovations ± 2 S.E. Response of LOG(PNHS) to LOG(NHS) .03
.02
.01
.00
-.01
-.02
-.03
1
2
3
4
5
6
7
8
9
10
11
12
Response to Cholesky One S.D. Innovations ± 2 S.E. Response of LOG(PNHSW) to LOG(NHSW) .04 .03 .02 .01 .00 -.01 -.02 -.03 -.04
1
2
3
4
5
6
7
8
9
10
11
12
35 Table 1. Growth Rates of Selected Series across U.S. Regions: 1983:I to 2006:I. (∆NHSi)/NHSi
(∆PNHSi)/PNHSi
(∆Yi)/Yi
(∆Y-i)/Y-i
Midwest
163.18%
73.64%
66.01%
86.22%
Northeast
-11.32%
230.01%
57.44%
88.48%
South
92.63%
97.54%
106.32%
70.17%
West
105.78%
218.78%
94.54%
77.71%
U.S.
91.49%
137.93%
81.31%
NA
Notes: Growth rates are calculated for each series for each region “i” between start and end of the sample. The first column has the growth rate of new home sales; the second has the growth rate of the price of new home sales; the third has the growth rate of local (own region) income; and the fourth column has the growth rate of aggregate income, defined as the sum of income of all other three states.
36 Table 2. Descriptive Statistics of U.S. Regions: 1983:I to 2006:I. Mean
Median
Maximum
Minimum
Std.Dev.
Skewness
Kurtosis
YPUS
4,104,215
3,937,620
5,306,917
2,922,919
687,135
0.176
1.782
CPIUS
147.56
149.40
199.80
97.90
29.30
-0.044
1.820
FF
5.500
5.500
11.300
0.980
2.544
0.032
2.342
R30
8.706
8.107
14.497
5.507
2.281
0.747
2.763
PNHS
140,032
130,000
247,700
73,300
43,130
0.623
2.839
NHS
787.43
718.00
1,297.00
463.33
203.82
0.910
3.082
YPUSNE
3,214,450
3,090,228
4,240,839
2,246,669
578,780
0.195
1.773
YPNE
889,765
855,947
1,066,078
675,393
109,473
0.060
1.894
CPINE
153.60
156.10
212.80
98.55
32.34
-0.062
1.899
PNHSNE
188,630
175,000
370,300
77,800
68,812
0.736
3.224
NHSNE
80.28
77.33
141.67
46.33
20.46
1.011
3.620
YPUSMW
3,140,555
2,991,465
4,126,830
2,212,430
551,826
0.210
1.794
YPMW
963,661
946,156
1,183,434
710,490
135,785
0.040
1.756
CPIMW
144.07
145.60
192.50
98.65
27.63
-0.027
1.759
PNHSMW
134,875
131,000
224,200
77,000
40,147
0.371
2.127
NHSMW
131.87
126.67
226.67
65.00
42.71
0.429
2.149
YPUSS
2,749,575
2,638,400
3,449,962
2,021,855
410,888
0.139
1.786
YPS
1,354,641
1,299,220
1,857,558
900,331
276,787
0.231
1.793
CPIS
144.00
145.80
192.80
98.40
27.19
-0.045
1.794
PNHSS
121,145
117,900
205,900
69,200
35,568
0.425
2.273
NHSS
357.41
329.00
662.33
194.33
109.42
0.979
3.330
YPUSW
3,198,870
3,082,268
4,083,309
2,297,543
514,743
0.151
1.780
YPW
905,345
855,353
1,223,608
625,316
172,618
0.253
1.804
CPIW
149.28
150.60
203.80
97.40
30.59
-0.017
1.827
PNHSW
164,729
145,000
344,300
78,800
65,267
1.052
3.420
NHSW
217.87
204.33
369.33
120.33
59.80
0.882
3.121
National
Northeast
Midwest
South
West
Notes: The total number of observations is 93. YPUS refers to aggregate income in the U.S. (in real USD 1,000s); YPUSi refers to aggregate income in the three regions other than “i” (in real USD 1,000s); YPi refers to local income of region “i” (in real USD 1,000s); CPIi to the national or regional price index (1982-1984 = 100); PNHSi to the price of new home sales of region “i” (in USD); and NHSi to the number of new home sales of region “i” (in 1,000s).
37 Table 3. Variance Decompositions of U.S. New Home Sales, 1983:I to 2006:I. Shocks in LOG (YPi)
Shocks in FF
Shocks in R30
Shocks in LOG (PNHSi)
Shocks in LOG (NHSi)
12.13
0.63
10.25
0.10
76.89
(6.15)
(2.06)
(5.32)
(1.22)
(7.49)
18.32
14.71
16.78
0.88
49.30
(8.88)
(8.76)
(7.35)
(2.36)
(9.57)
24.42
21.13
15.14
5.96
33.35
(10.88)
(12.40)
(8.01)
(7.03)
(10.47)
6.75
0.01
0.59
0.58
0.04
92.03
(5.17)
(1.58)
(2.11)
(2.06)
(1.35)
(5.85)
5.02
1.06
1.52
1.86
2.00
88.54
(5.22)
(3.38)
(4.62)
(4.42)
(4.65)
(8.68)
4.02
1.76
3.19
5.56
7.43
78.05
(4.52)
(5.17)
(8.60)
(8.60)
(9.16)
(13.01)
7.08
2.28
0.02
2.15
0.80
87.67
(5.23)
(3.11)
(1.48)
(3.07)
(2.00)
(6.49)
17.24
12.08
7.97
10.84
2.06
49.81
(7.93)
(7.36)
(5.76)
(5.50)
(2.99)
(7.73)
17.11
16.71
15.35
9.12
1.70
40.01
(8.77)
(9.64)
(9.45)
(5.15)
(3.79)
(8.28)
5.13
0.07
2.76
3.31
0.36
88.38
(4.72)
(1.48)
(3.49)
(3.60)
(1.65)
(6.51)
8.86
8.88
4.24
8.74
4.16
65.12
(7.40)
(6.21)
(5.85)
(4.79)
(4.19)
(9.92)
6.01
21.95
6.05
12.23
4.10
49.67
(6.17)
(10.98)
(8.33)
(6.86)
(6.69)
(11.72)
6.71
1.55
1.72
12.39
4.05
73.59
(5.02)
(2.69)
(2.77)
(6.29)
(3.59)
(7.90)
14.31
4.26
10.42
16.04
4.66
50.31
(8.10)
(4.80)
(6.85)
(6.63)
(4.44)
(9.38)
14.43
3.53
25.09
12.53
6.14
38.28
(8.41)
(6.40)
(10.83)
(6.22)
(4.98)
(9.53)
Shocks in LOG (YP-i) National 1 quarter
6 quarters
12 quarters
Northeast 1 quarter
6 quarters
12 quarters
Midwest 1 quarter
6 quarters
12 quarters
South 1 quarter
6 quarters
12 quarters
West 1 quarter
6 quarters
12 quarters
Notes: Standard deviations calculated by 1,000 Monte Carlo Simulations are in parenthesis. The underlying model is either a 5variable national VAR or a 6-variable regional VAR. Similar decompositions are found for the 7-variable VAR with inflation.
