Does housing really lead the business cycle Luis

Does housing really lead the business cycle? Luis J. Álvarez and A. Cabrero Conference on Macroeconomics of Housing Markets Banque de France, 3 -4 December 2009 DG ECONOMICS, STATISTICS AND RESEARCH

Outline 1. Motivation 2. Methodologies for estimating cycles 3. The leading nature of housing 4. Brevity of contractions 5. Conclusions DG ECONOMICS, STATISTICS AND RESEARCH 2

Motivation █ 1996 -1998 █ 2005 -2007 § The analysis of housing cycles in Spain is particularly relevant –Strong investment in residential construction in the decade prior to 2006 • low interest rates and sizable migration inflows. • Housing share in GDP substantially above that in the euro area or the US • Highly beneficial impact on employment –soaring house prices • … despite the marked expansion of housing supply § Leamer (2007) shows for the US economy that residential investment offers the best early warning sign of an oncoming recession among GDP components § The natural question that arises is whether this also holds true for Spain DG ECONOMICS, STATISTICS AND RESEARCH 3

Measurement of cycles § There is a wide variety of procedures to estimate business cycles § Different methods involve a different concept of the cycle, so it is often misleading to compare across methods § We define the business cycle as the outcome of an ideal bandpass filter § This concept is in contrast with the deviation of output from output consistent with stable inflation (production function approach) output of an economy without nominal rigidities (DSGE models) DG ECONOMICS, STATISTICS AND RESEARCH 4

The ideal band pass filter § The filter fully removes high-frequency fluctuations (e. g. those with a period of less than 6 quarters (1. 5 years)) … and also long-run movements (e. g. over 32 quarters (8 years)) The method is flexible to accomodate other definitions Medium term cycles (Comin and Gertler (2006) § The gain function of a filter indicates the extent to which it affects the series § A gain over 1 indicates that those fluctuations are amplified § A gain of 1 implies no effect § A gain of 0 shows that fluctuations are fully suppressed DG ECONOMICS, STATISTICS AND RESEARCH 5

Polynomial regression (kernels) § Popularised by Leamer(2007) for business cycle analysis § Well known method in the statistical literature (Stone (1977), Cleveland (1979)) § The method requires to estimate a weighted least square regression for each date. Variants differ in the weights (kernel) used, number of dates involved in the local regression (bandwidth) and order of the time polynomial § The method is equivalent to a certain two-sided moving average, which allows a frequency domain interpretation Álvarez and Cabrero (2009) § Some insatisfactory properties – Cyclical fluctuations are dampened – Long run movements are not suppressed DG ECONOMICS, STATISTICS AND RESEARCH 6

Butterworth filters § Widely used in engineering in their onesided form (Butterworth(1930)) § Can be seen as a generalization of the HP filter (The HP filter is a particular lowpass Butterworth filter of the sine) § Highly flexible. Can be given a modelbased interpretation § Filters are very close to the ideal bandpass filter – These filters are able to remove satisfactorily short-run and medium run fluctuations DG ECONOMICS, STATISTICS AND RESEARCH 7

Alternative filters § The standard HP filter is a low-pass filter, so that it does not remove short-run fluctuations (and passes through substantial long-run movements) § A bandpass HP filter (i. e. the difference of 2 HP filters) provides a better approximation, but still passes through substantial long-run movements §Baxter and King (1999) propose a moving average approximation to the ideal filter. Their method involves losing k observations at the end (the most interesting period for policy-makers!) and beginning of the series. Unless the moving average is very high, properties are not satisfactory § Christiano and Fitzgerald (2003) derive a filter which is optimal under the assumption of random walk behaviour (not very realistic for GDP!) DG ECONOMICS, STATISTICS AND RESEARCH 8

Comparisons of filters GDP § From a theoretical point of view, Butterworth filters are clearly to be preferred § In practice, differences are important in terms of volatility, but not much in terms of developments –Different estimates are highly correlated, except for house prices Housing investment § We use series extended with forecasts to obtain a more accurate estimation of the cycle at the end of the sample § Our starting point is not seasonally adjusted series, but trend series. This allows to obtain clearer signals and minimise problems of some filters with highfrequency fluctuations DG ECONOMICS, STATISTICS AND RESEARCH 9

The leading nature of housing (I) § Residential investment leads GDP: current residential investment is highly correlated with future GDP § Non residential construction lags GDP § Fluctuations in the different investment items much higher than those in GDP DG ECONOMICS, STATISTICS AND RESEARCH 10

The leading nature of housing (II) § All filters consistently show that housing investment leads GDP § This is a challenge for theoretical models, which generally do not account for this fact –There are some attempts to explain the lead of residential vs non-residential investment • In Fisher (2007) residential investment leads fixed investment, but not GDP • Yuan (2009) presents a model in which residential investment leads GDP. Agents face collateral constraints and receive signals about future productivity –A good signal about future productivity makes households spend more to intertemporally smooth consumption. Increased expenditures are financed by borrowing at mortgage interest rates. As a result, agents buy more housing relative to other goods. – Is it realistic to assume that households continuously vary the size of mortgages, according to fluctuations in total spending? DG ECONOMICS, STATISTICS AND RESEARCH 11

The leading nature of housing (III) Housing starts and building permits clearly lead GDP: it takes time to build a house Construction decisions by firms and the public sector do not lead GDP GVA in construction comoves with GDP Investment in non residential construction lags GDP Labour market variables also lead GDP DG ECONOMICS, STATISTICS AND RESEARCH 12

Demand versus supply shocks § House prices consistently comove with housing investment This is consistent with a more relevant role of demand factors, such as demographics and interest rates, over supply ones, such as technical progress in the housing sector. . . but land use constraints probably have also played an important role § House prices lag residential investment This probably reflects house price stickiness DG ECONOMICS, STATISTICS AND RESEARCH 13

Asymmetries in expansions and contractions Durations in quarters § GDP contractions tend to be briefer than GDP expansions Capacity constraints (Hansen and Prescott (2005)), Credit constraints (Kocherlakota (2000)) § Contractions in residential investment and other real indicators are also briefer than expansions § There also assymetries in the labour market Mc. Kay and Reis (2008) § The response of prices is quite symmetric DG ECONOMICS, STATISTICS AND RESEARCH 14

Assymmetries in leads and lags § The lead of residential investement over GDP is smaller in contractions than in expansions § For the rest of variables, there are no clear patterns. Results depend substantially on the filter Caveat: Our sample starts in 1980, so there is a small number of turning points DG ECONOMICS, STATISTICS AND RESEARCH 15

Conclusions (Main facts) 1. Main result: Residential investment and, more clearly, housing starts lead GDP Leamer(2007) Challenge for theoretical work 2. House prices consistently comove with housing investment More relevant role of demand factors, such as demographics and interest rates, over supply ones, such as technical progress in the housing sector. . . but land use constraints probably have also played an important role 3. House prices lag residential investment This probably reflects house price stickiness 4. GDP contractions tend to be briefer than GDP expansions Also for residential investment and other construction real indicators DG ECONOMICS, STATISTICS AND RESEARCH 16