Reconstructing the Great Recession

Abstract

This article uses dynamic equilibrium input-output models to evaluate the contribution of the construction sector to the Great Recession and the expansion preceding it. Through production interlinkages and demand complementarities, shifts in housing demand can propagate to other economic sectors and generate a large and sustained aggregate cycle. According to our model, the housing boom (2002-07) fueled more than 60 percent and 25 percent of employment and GDP growth, respectively. The decline in the construction sector (2007-10) generates a drop in total employment and output about half of that observed in the data. In sharp contrast, ignoring interlinkages or demand complementarities eliminates the contribution of the construction sector. 

Introduction

With the onset of the Great Recession, U.S. employment and gross domestic product (GDP) fell dramatically and then took a long time to return to their historical trends. There is still no consensus about what exactly made the recession so deep and the subsequent recovery so slow. In this article we evaluate the role played by the construction sector in driving the boom and bust of the U.S. economy during 2001-13.

The construction sector represents around 5 percent of total employment, and its share of GDP is about 4.5 percent. Mechanically, the macroeconomic impact of a shock to the construction sector should be limited by these figures; we claim it is not. Rather, we claim one reason why the Great Recession was particularly deep and persistent is that the construction sector and housing consumption are strongly interconnected to the rest of the economy. Such linkages are important at the production stage (purchases of intermediate goods) but also at the final consumption stage (broad demand complementarities).

Our vision of how a shock to the demand for housing travels to the rest of the economy is the following. In response to a demand-driven housing boom, the construction sector leads the rest of the economy by fueling an expansion through its purchases of inputs. The expansion generates an increase in consumption (housing and nonhousing) as well as investment (residential and capital). This continues either until a new steady state is reached or, as in the historical case we study, a sudden drop in housing demand generates a decline in construction output. This translates into a general reduction of demand (for intermediate inputs and complementary consumption goods), thereby propagating and magnifying, again, the negative sectoral shock. Further, a sudden drop in the demand for housing also generates a slow recovery because the excessive inventory of housing units takes a long time to be absorbed, hence the particularly long delay in the aggregate recovery.

In the empirical analysis, we construct measures of sectoral interlinkages (multipliers) and show that the construction sector is one of the most interconnected in the economy. We use this to quantify the contribution of the construction sector during the period 2002-13 and estimate it to have been unusually large. Construction is capable of accounting for about 52 percent of the decline in total employment and 35 percent of the decline in aggregate gross output. Eliminating the production multipliers weakens the impact on total employment to 20.8 percent and on gross output to 19.3 percent.

In a simplified version of the model, we illustrate the importance of production interlinkages and demand complementarities. This exercise provides a set of sufficient conditions under which the presence of interlinkages generates larger effects in aggregate employment and output than in their absence. The algebra indicates that, for the amplification effect to exist, the sectoral interlinkages must be asymmetric, with the construction sector buying relatively more inputs from the rest of the economy than vice versa. This condition is supported—by more than two orders of magnitude—by estimates from the U.S. input-output table from the Bureau of Economic Analysis (BEA). To generate a multiplier effect via interlinkages, it is also necessary to have an elasticity of substitution between housing and consumption goods lower than 1. With an elasticity of substitution larger than 1, a decline in housing demand generates the reallocation of productive inputs away from the construction sector and a boom in the nonhousing sectors, which may (more than) compensate for the decline due to the supply-side interlinkages. With a unitary elasticity, these two effects cancel out.

One of the limitations of the static model is that it does not allow study of the dynamic adjustment of consumption, residential investment, and productive capital. It also ignores the process of the adjustment of relative prices, and it is not ideal for quantitative purposes. To overcome these limitations, we solve the full dynamic general-equilibrium model numerically. We use that model to answer the following question: If demand for housing shifts exogenously over time to match the observed dynamics of employment in the construction setor, what will happen to the remaining macro quantities and prices?

Our simulations indicate that, in the presence of sectoral interlinkages and consumption complementarities, the size of the boom-bust cycle in total employment and output is substantially larger and more coherent with historical observations than otherwise. During the demand-driven housing boom, both sectors in our model expand and contribute to the growth of output and employment, by 2 percent and 2.5 percent, respectively. During the housing bust, the decline in output is 3.3 percent and in total employment is 3.8 percent. The model also captures the leading role of construction during booms and busts (see Leamer, 2007) and the comovement with nonhousing expenditures and investment. Further, the separation of productive capital and residential structures, together with the irreversibility constraints, introduces an asymmetry between expansions and recessions similar to that observed by many previous researchers and that is traditionally hard to obtain in most real business cycle models. As in the theoretical exercise, the dynamic quantitative simulations show that reducing the importance of either the sectoral interlinkages or the demand complementarities weakens the transmission mechanism. Moreover, the model without linkages also fails to capture the lead-lag pattern of housing and consumption expenditures observed in the data. These results indicate that modeling production linkages provides a quantitatively relevant transmission channel.

The burst of the real estate "bubble" might have substantially lowered potential output and created a substantial "displacement effect," for both labor and capital, which took quite some time to absorb. Some researchers have referred to this displacement effect as a worsening of labor frictions. For example, Arellano, Bai, and Kehoe (2019) and Ohanian and Raffo (2012) attribute the Great Recession primarily to this factor. Since our model captures significant declines in employment and output in the absence of such frictions, we also perform a business cycle accounting exercise on simulated data from the model. Through the lens of the one-sector neoclassical growth model, the presence of intersectoral linkages, movements in relative prices, and shifts in housing demand can be interpreted as "distortions." Business cycle accounting would attribute the recession generated in the model to the labor wedge. In our model, the magnitude of the worsening of the labor wedge is about 62 percent of the total change observed in the data. Importantly, in both our model and the data, the worsening is due to the consumer side of the labor wedge and not to differences between wages and the marginal product of labor.

Obviously, the fluctuations of the construction sector cannot fully account for the dynamics of employment and output since 2002. Other relevant factors not incorporated into the analysis are important. Many suggest (Black, 1995; Hall, 2011; and Kocherlakota, 2012) that high interest rates could be responsible for the slow recovery. These authors argue that even in models with perfect competition and price flexibility (i.e., lacking the typical frictions of New-Keynesian business cycle models), too-high interest rates may result in substantially lower levels of output and employment. Since some interest rates appear to be currently constrained by the zero lower bound, such analyses appear particularly pertinent. Others argue that the level of uncertainty (Bloom, 2009, and Arellano, Bai, and Kehoe, 2019), government policies (Herkenhoff and Ohanian, 2011), and excessive debt overhang in the economy (Garriga, Manuelli, and Peralta-Alva, 2019; Herkenhoff and Ohanian, 2012; and Kehoe, Ruhl, and Steinberg, 2013) may be responsible for the lackluster recovery. Our exercise is silent with respect to these factors.