![Sample our Built Environment journals, sign in here to start your access, Latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5926/TOC-Subject-Sample-2021-Built-Environment.jpg"})
Abstract
We use the price and quantity of credit to distinguish credit supply shocks from credit demand shocks to examine international housing returns around the financial crisis. We find little evidence to support three popular credit supply explanations for the run-up in international housing prices before the financial crisis. Importantly, we show that common credit supply proxies often represent credit demand shocks because they only capture quantity supplied but ignore price. Furthermore, credit demand shocks are more common than supply shocks in countries that experience a housing reversal. While we do find that credit supply plays a role in the housing run-up, the impact of credit supply shocks on housing prices during the run-up is primarily through negative shocks driving prices down, rather than the positive supply shocks driving them up.
Notes
1 Though the use of LTV as a source of credit demand shocks is common in these structural models, Glaeser et al. (Citation2013) find that the median LTV ratio in the U.S. in 2005 was no higher than the LTV in 1999, and their model suggests that LTV shocks to housing demand can’t account for the observed price changes in the U.S. housing market. Justiniano et al. (Citation2015b) and Iacoviello and Neri (Citation2010) also show that shocks to the LTV ratio alone have only a small impact on housing prices.
2 Favilukis et al. (Citation2013) use the error from a regression of the credit standards onto the credit demanded. Note, however, that the credit demand curve must be stable in order for this orthogonalized variable to proxy for movements in the credit supply curve.
3 Cerutti et al. (Citation2017) document that house-price booms are linked to credit booms in both household and corporate credit. They measure household credit using BIS data on household credit and corporate credit (which gives total credit) to the private sector.
4 As Cesa-Bianchi et al. (Citation2018) point out, rising asset prices can amplify the supply shock through inflated collateral values. They develop a model where exogenous changes in financial intermediaries’ leverage generates credit supply shocks. They empirically show that increases in the leverage of U.S. broker-dealers are associated with increases in house prices and cross- border credit flows, consistent with a positive credit supply shock.
5 Eurostat (Citation2013) provides good overviews of the problems in constructing a house price index, and Hill and Mesler (Citation2008) provide a technical review of the problems in constructing price indexes in general. While these studies note the “correct” or best price index methodology is not universal and dependent upon the data available, Bourassa et al. (Citation2006) and Nagaraja et al. (Citation2014) show that different methodologies often result in indexes that are highly correlated. This correlation is quite strong at annual horizons but is weaker at quarterly horizons and for indexes based on an unadjusted median.
6 Housing data for Mexico starts after 2002Q4, hence Mexico is absent from the graph.
7 Tables containing the number of observations on housing prices by country, as well as the number of credit demand and credit supply shocks identified for each country based on the proxies for the price and quantity of credit, are available from the authors upon request.
8 We do, however, restrict the equilibrium real rate to lie between 1.5 and 8 percent.
9 Including policy inertia and a time-varying inflation target provides the best overall fit in explaining housing returns. The correlation between the different specifications of the Taylor rule, however, ranges between 0.81 and 0.99, and the regression results are quite similar under all specifications.
10 Graphs of the number of countries with overly tight/loose monetary policy, as well as the number of countries with credit supply shocks that coincide with the policy stance, are available from the authors.
11 The one exception is Canada’s survey on credit supply which asks about business lending but not mortgage or household loans.
12 Favilukis et al. (Citation2013) use weights for prime loans of 0.75 in 2007 and 0.95 in 2008, which are the only years in their data that have subprime loans. Subprime loans appear again in the survey beginning in 2012Q2. We continue to use a weight of 0.95 for prime loans during this period.
13 In addition, some countries report the diffusion index as a positive value for easing credit standards and negative for tightening. To keep our data consistent, we use the negative of the given diffusion index for such countries. For each country positive values indicate that credit standards are stronger and negative values indicate weaker standards.
14 There is also evidence the survey data is correlated with actual changes in price and quantity. Regressing changes in the price (quantity) of credit next quarter onto changes in the credit demand survey results in a positive slope coefficient that is significant at the 5% (10%) level. Regressing changes in the quantity of credit next quarter onto changes in the credit standards survey results in a negative and significant slope coefficient at the 5% level.
15 Note that these credit demand and credit supply shocks identified with central bank surveys on mortgage and household loans do not need to coincide with the shocks identified using total real credit to the non-financial sector and non-central bank depository institutions’ lending rate. While discrepancies do occur, they are rare. Furthermore, dropping observations where there is a conflict results in only small changes in average housing returns, and does not change the qualitative results in Table 2 or any of the subsequent tables.
16 We start the pre-crisis period in 2002Q4 since we evaluate cumulative housing price changes relative to all three of our credit supply proxies, and most central banks don’t administer credit surveys prior to 2002Q4. We end the post-crisis period in 2012Q4 so that the pre-crisis and post-crisis samples are of equal length.
17 As is common in the literature (e.g. Guerrieri and Iacoviello (Citation2017)), for identification we assume that the housing supply is constant so that housing price movements are attributable to housing demand shocks.
18 Housing data for Mexico don’t begin until 2005 so Mexico is excluded from the pre-crisis regression, resulting in 53 observations.
19 Countries must have at least four years of survey data to be included. Columbia, Denmark, Indonesia, Latvia, Malta, Norway, Philippines, Slovakia, Slovenia, Thailand, and the U.K. are not included in the pre-crisis regression due to the start date of the survey. The Czech Republic, Estonia, and Russia are left out of both of these regressions due to the late start date of the survey data. In the pre-crisis regression we include Poland, and in the post-crisis regression Slovakia, Philippines, and Indonesia despite having only 4-years worth of survey data and not five.
20 The fixed effects in our panel models will capture elasticity at the national level as long as national elasticity is stable. This seems plausible given nations’ supply elasticities are a weighted average of their cities’ elasticities, and Saiz (Citation2010) finds such elasticities to be stable over long time horizons. Nevertheless, stability in supply elasticity is less likely true for nations than for cities.
21 While local elasticities are often assumed to be stable over long periods of time, it is unclear if the same is true of national elasticities despite the fact they are themselves weighted averages of local elasticities. The results, however, will not be materially affected unless the changes in elasticity result in a substantial recategorization of countries between high and low elasticity. That is, the grouping of the elasticity estimates needs to be stable, not the elasticity estimates themselves. Furthermore, we only need the grouping to be stable over the relatively short, 5-year pre-crisis period.
22 Matches are made two ways: (1) the countries’ predicted pre-crisis housing return and (2) the predicted number of housing starts (data is collected form the OECD). Once matched, the country with the lower actual return (higher housing starts) is assigned to the high elasticity group. Both methods provide the same results, and the predictive models include the population growth, population density, real consumption, and real GDP as control variables. Details and tabulated results are available upon request.