Sam Christie and Aniruddha Rajan
Sudden contractions in credit supply can trigger and amplify recessions – a reality made painfully clear by the 2008 global financial crisis (GFC). However, quantifying these real economic effects is challenging. In this post, we demonstrate a novel way to do so using Granular Instrumental Variables (GIV), focusing on the UK mortgage market. The core idea is that we can exploit the market’s concentration to build up exogenous fluctuations in aggregate credit supply from idiosyncratic lender-specific shocks. Using our GIV, we find evidence that contractionary mortgage supply shocks can have quantitatively significant effects on the macroeconomy, causing persistent decreases in output, consumption, and investment, alongside increases in unemployment.
Why bother with an instrumental variable?
To understand the impact of credit supply shocks, we could try simply regressing some macroeconomic aggregates on a measure of credit volumes. However, simultaneous causality invalidates this approach – we may erroneously pick up changes in credit supply that are actually driven by the business cycle. A recession, for example, could tighten lenders’ balance sheet constraints (directly reducing credit supply) or decrease credit demand to which lenders respond (indirectly reducing credit supply).
A solution is to use an instrumental variable to identify variation in credit supply unrelated to macroeconomic conditions. But unfortunately such instruments are hard to find at the aggregate level. Our work uses a novel form of instrument to overcome this challenge – a GIV. Our GIV approach relies on two features of credit markets: (i) high market concentration and (ii) sufficiently volatile idiosyncratic shocks to lenders. Markets of this kind are called granular (Gabaix (2011)). If mortgage lenders display granularity, idiosyncratic shocks to them shouldn’t wash out at the aggregate level, generating exogenous movements in market outcomes – a valid instrumental variable.
Why focus on the mortgage market?
The mortgage market is a particularly relevant credit market because mortgages constitute the largest liability of a typical UK household (BIS (2023)). This makes mortgage debt a key determinant of real disposable income so shocks to its supply can, in principle, impact the macroeconomy. For example, a contraction in mortgage lending could reduce consumption by lowering house prices and reduce investment by slowing new housing construction. These real economic spillovers mean that mortgage supply shocks are relevant for policy, warranting close attention from the Bank of England’s policymaking committees.
Additionally, the UK mortgage market is highly concentrated, making it a strong candidate for a GIV analysis. Figure 1 illustrates the concentration of the market via a Lorenz curve of lender market shares. The dominance of the so-called Big-6 lenders causes the Lorenz curve to deviate substantially from the 45-degree line of perfectly equal lender size, as reflected in a high Gini coefficient (67%) and ‘Big’-6-firm concentration ratio (69%). This concentration is promising for using GIV but we also need idiosyncratic shocks to mortgage lenders to be sufficiently volatile. Fortunately, other work in the banking literature indicates the latter is true, with lender-specific events such as unexpected loan provisions, capital injections, and cyber problems cited as regular occurrences in credit markets. These are exactly the types of idiosyncratic shocks that we’re trying to capture in our GIV.
Figure 1: Lorenz curve for the mortgage market
Notes: Lorenz curve, Gini coefficient and ‘Big’-6-firm concentration ratio for the UK mortgage market. Calculations are made as averages across the sample period. Each blue dot along the Lorenz curve represents a lender. The black dashed line is a theoretical Lorenz curve where all lenders are equally sized.
How do we construct our GIV?
We construct our GIV using lender-level data on the stock of mortgages issued by monetary financial institutions at a quarterly frequency. The Bank of England collects these confidential data and publishes the aggregated series on Bankstats. Our sample focuses on the period from 2010 Q1 to 2019 Q4 for two reasons: (i) lender-specific data before the GFC are limited and (ii) the volatility in the data created by Covid-19 represent a structural break that we want to avoid.
With our lender-level data, we follow Gabaix and Koijen (2024) to isolate idiosyncratic mortgage supply shocks using a combination of parametric and non-parametric techniques. Among other elements, this involves actively controlling for lenders’ mortgage demand using the Bank of England’s Credit Conditions Survey to ensure that the variation we identify truly represents supply-side changes (as done in Financial Stability Paper No. 51). We then form our aggregate GIV by size-weighting and summing the idiosyncratic shocks across lenders. Intuitively, the size-weighting means that large lenders gain more importance in our final instrument. These institutions should have greater influence on market outcomes so this process ensures our GIV will be as relevant an instrument as possible.
Why bother with a PCA?
To increase our confidence that we capture truly idiosyncratic movements in mortgage supply, we take a further step to clean our shock series by performing a principal component analysis (PCA). The PCA allows us to strip away variation in mortgage supply that may still be related to macroeconomic conditions but to which lenders have differing sensitivities. For example, heterogeneity across lenders’ risk appetites could mean they respond differently to the business cycle. Figure 2 illustrates the value of this additional step by comparing correlations between the lender-specific shocks we extract before and after the PCA. The correlations between lender-specific shocks become much closer to zero (ie uncorrelated) after the PCA, suggesting these shocks are indeed now idiosyncratic.
Figure 2: Impact of a PCA on correlations between lender-specific shocks
Notes: Correlation matrices between lender-specific mortgage supply shocks in our GIV. The left-hand matrix is when we do not perform a PCA on the lender-specific shocks and the right-hand matrix is when we do (extracting two principal components).
What are our key results?
Our first key result is that the UK mortgage market is granular. Our GIV is a strong instrument for aggregate mortgage volumes (with an F-statistic well above 10), confirming that idiosyncratic shocks to lenders can explain movements in the mortgage market as a whole. Given the difficulty in finding valid macroeconomic instruments, this is a non-trivial finding and validates the use of our GIV in this setting.
Our second key result is that aggregate contractions in mortgage supply can indeed have substantial real effects on the macroeconomy. Using our GIV in a local projection, we trace the impact of a mortgage supply shock on different macroeconomic variables. Figure 3 displays our baseline estimates for a one standard deviation contraction in mortgage lending (67 basis points). The shock causes statistically significant decreases in output, consumption, and investment, which persist over time. Real output falls by 1.3 percentage points after two years, which is underpinned by peak falls in consumption and investment of 1.3 percentage points and 3.0 percentage points, respectively. The shock also causes a peak rise in the unemployment rate of 0.3 percentage points, though the latter effect is more muted. For comparison, the largest quarterly movement during the global financial crisis amounted to a 1.6 standard deviation contraction in mortgage lending. This suggests that, while only one of many factors, large mortgage supply shocks can have quantitively significant effects on the UK economy.
Figure 3: Response of macroeconomic aggregates to a negative mortgage supply shock
Notes: Cumulative response of output, consumption, investment, and unemployment to a one standard deviation contraction in mortgage supply. Estimated via local projection using a GIV constructed after extracting two principal components. 68% confidence bands in orange and 95% confidence bands shaded grey, with Newey-West standard errors (four lags). Local projection controls for four lags of the dependent variable. Note that investment is measured by total gross fixed capital formation.
As a reference point and to illustrate the value of the instrument, we run the same specifications using standard OLS. OLS generates less persistent responses of all variables to the mortgage supply shock. The responses are also estimated less precisely, with the confidence bands for the macroeconomic responses regularly crossing zero. This imprecision is particularly prominent for the unemployment rate.
Our GIV results are qualitatively similar to those obtained elsewhere in the literature using alternative methods. This includes Barnett and Thomas (2014) who identify credit supply shocks in the UK using standard macro-econometric techniques on aggregated data. Our analysis builds on their efforts by leveraging micro-econometric techniques on disaggregated data to provide stronger identification. Hence, our GIV work provides novel empirical support to the vast theoretical literature on how credit supply shocks can cause and amplify recessions (see Kiyotaki and Moore (1997) and Diamond and Rajan (2005)).
Our finding that credit supply shocks can have real economic effects is also relevant for policymaking. From a monetary policy perspective, it demonstrates that monitoring these shocks and understanding their transmission are salient issues to the Bank of England’s Monetary Policy Committee. There are implications for financial stability too because systemic risks can trigger shocks to aggregate credit supply (as experienced during the GFC). In terms of macroprudential regulation, this underscores the importance of the Financial Policy Committee in identifying, monitoring, and acting against such risks to the financial system. Our results additionally justify the use of microprudential regulation by the Prudential Regulation Committee to ensure that lenders are sufficiently capitalised in the face of credit supply shocks. This is regardless of whether these shocks are system-wide or idiosyncratic, as we show that even the latter can affect aggregate outcomes given the granularity of the UK banking system.
What are the key takeaways?
First, we show that a novel method of constructing instrumental variables (GIV) is applicable to credit markets. Specifically, we apply this approach to the highly concentrated UK mortgage market and find that idiosyncratic supply shocks can influence aggregate outcomes. Obtaining valid macroeconomic instruments is challenging but we demonstrate that leveraging micro-data in a suitably granular market can address this issue.
Second, contractionary shocks to mortgage supply can have material real economic impacts. A reduction in mortgage supply decreases output, consumption, and investment, and increases unemployment. These effects are persistent and economically significant, suggesting that when credit supply shocks do occur they can drive business cycles. This provides motivation for the Bank of England’s policy committees to understand, monitor and act against such shocks. In other words, credit supply shocks matter!
Sam Christie and Aniruddha Rajan work in the Bank’s Monetary and Financial Conditions Division.
If you want to get in touch, please email us at bankunderground@bankofengland.co.uk or leave a comment below.
Comments will only appear once approved by a moderator, and are only published where a full name is supplied. Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.
Share the post “GIV us some credit: estimating the macroeconomic effects of credit supply shocks”
