Household Debt and Monetary Policy: Revealing the Cash Flow Channel By Flodén, Kilström, Sigurdsson, Vestman Discussion: Daniel L. Greenwald (MIT Sloan) Econometric Society Meetings January 7, 2016 Daniel L. Greenwald (MIT Sloan) Household Debt and Monetary Policy January 7, 2016 1 / 8
Summary Main question: how much of monetary policy transmission into consumption is due to changes in current cash flows? - Contrast with intertemporal substitution channel. Approach: use cross-sectional wealth and consumption data to isolate variation in mortgage indexation. - Also use variation in debt-to-income (DTI) ratio. - Redo estimates using IV, identified monetary policy shocks for robustness. Main findings: strong transmission through cash flow channel. - Consumption response increasing in indexation, DTI ratio. - Large implied MPCs out of cash flows. My take: careful empirical exercise using great data. - But may not have completely isolated cash flow channel due to principal repayment decision. Daniel L. Greenwald (MIT Sloan) Household Debt and Monetary Policy January 7, 2016 2 / 8
Simple Model (Exogenous Payment) Consider simplified deterministic model with CRRA utility. - Only control variable is bond b i,t, exogenous process for debt d i,t. Constraints: c i,t y i,t + b i,t 1 Rt 1 b i,t r d i,t d i,t 1 + d i,t b i,t 0. Assuming CRRA preferences (EIS σ): log c i,t+1 σ log β + σ log R t. Interior solution: c i,t moves with R t due to intertemporal substitution. - Mortgage payment only influences c i,t via permanent income/wealth. Corner solution: c i,t = y i,t + b i,t 1 r d i,t d i,t 1 + d i,t. - c i,t moves with payment (cash-flow channel), no direct effect of R t. Daniel L. Greenwald (MIT Sloan) Household Debt and Monetary Policy January 7, 2016 3 / 8
Simple Model (Exogenous Payment) Consider simplified deterministic model with CRRA utility. - Only control variable is bond b i,t, exogenous process for debt d i,t. Constraints: c i,t y i,t + b i,t 1 Rt 1 b i,t r d i,t d i,t 1 + d i,t b i,t 0. Assuming CRRA preferences (EIS σ): log c i,t+1 σ log β + σ log R t. Interior solution: c i,t moves with R t due to intertemporal substitution. - Mortgage payment only influences c i,t via permanent income/wealth. Corner solution: c i,t = y i,t + b i,t 1 r d i,t d i,t 1 + d i,t. - c i,t moves with payment (cash-flow channel), no direct effect of R t. Daniel L. Greenwald (MIT Sloan) Household Debt and Monetary Policy January 7, 2016 3 / 8
Simple Model (Exogenous Payment) Consider simplified deterministic model with CRRA utility. - Only control variable is bond b i,t, exogenous process for debt d i,t. Constraints: c i,t y i,t + b i,t 1 Rt 1 b i,t r d i,t d i,t 1 + d i,t b i,t 0. Assuming CRRA preferences (EIS σ): log c i,t+1 σ log β + σ log R t. Interior solution: c i,t moves with R t due to intertemporal substitution. - Mortgage payment only influences c i,t via permanent income/wealth. Corner solution: c i,t = y i,t + b i,t 1 r d i,t d i,t 1 + d i,t. - c i,t moves with payment (cash-flow channel), no direct effect of R t. Daniel L. Greenwald (MIT Sloan) Household Debt and Monetary Policy January 7, 2016 3 / 8
Empirical Approach Authors limit variation to mortgage indexation r d i,t (R t), control for interest rate, individual, time fixed effects. - Should kill variation in intertemporal substitution for households at interior solution. Also controls for correlation between y i,t and R t, as long as not too much selection in mortgage indexation choice. - Should also control for many other effects in richer model (house prices, credit constraints, fiscal response, etc). Cash-flow channel should survive, impact on log c i,t approx. proportional to DTI i,t 1 r d i,t. Daniel L. Greenwald (MIT Sloan) Household Debt and Monetary Policy January 7, 2016 4 / 8
Simple Model (Endogenous Payment) Complication emerges when borrower can choose how much principal to repay. HH chooses mortgage payment x i,t (newly issued debt d + i,t exogenous): c i,t y i,t + b i,t 1 Rt 1 b i,t x i,t + d + i,t b i,t 0 x i,t (r d i,t + ν i)d i,t 1 Optimality condition for x i,t : d i,t = R d i,t d i,t 1 x i,t + d + i,t. log c i,t+1 σ log β + σ log R d i,t. Challenge: isolating difference in R d i,t due to indexation cannot kill intertemporal substitution effect for interior debt solution. Daniel L. Greenwald (MIT Sloan) Household Debt and Monetary Policy January 7, 2016 5 / 8
Recovering Cash Flow Channel Intuition: ability to pay principal on mortgage like special bond that only you can invest in. Interior solutions particularly likely with Swedish data because no-amortization loans are common (56% of loans in 2011 but falling!). To re-isolate cash flow channel, drop individuals who make more than the minimum payment (reverse of Vissing-Jorgenson, 2002). - Should be possible given data. - However, may introduce new selection issues (e.g., amortization type). Daniel L. Greenwald (MIT Sloan) Household Debt and Monetary Policy January 7, 2016 6 / 8
Amortization Appear to be population 0 differences across amortization types. FINANSINSPEKTIONEN THE SWEDISH MORTGAGE MARKET 14. AMORTISATION AS A SHARE 2011 2012 2013 2014 2015 Effects of FI s proposed amortisatio 13. AMORTISATION AS A SHARE OF OF THE corresponding LOAN AMOUNT figure FOR in 2014 was 3.2 15. per AMORTISING cent and HOUSEHOLDS in 2011 1.8 per cent. THE LOAN AMOUNT FOR DIFFERENT DIFFERENT The higher DEBT-TO-INCOME loan-to-value ratio a household FOR VARIOUS In has, December AGE the GROUPS more 2015, they FI amortise submitted a proposal for an am- 1 LOAN-TO-VALUE RATIOS (Per cent) RATIOS (Per cent) (Share of households, per cent) 4 5 in relation to their income. 100 ortisation requirement for consultation. debt-to-income ratios to 14 In brief, the proposal It is more common for households with low 4 amortise, and average amortisation in relation to the size of the loan is 80 entailed that new mortgages must be 3 higher for these households (Diagram 14). Amortisation amortised by 2 in per relation cent of the to total loans decreases when debt-to-income increases. loan amount However, annually households if the loanto-value portion ratio of exceeds their income 70 per cent, 3 60 with high debt-to-income ratios place a larger 2 on amortisation than households with lower and debt-to-income by 1 per cent of ratios. the total loan 13 2 Younger households amortise to 40 a greater extent amount than annually older if households the loan-to-value (Diagram 15). This is probably mainly because ratio they is between also have 50 and higher 70 per cent. 1 loan-to-value ratios. However, amortisation In has FI s increased consultation to more memorandum, 1 20 less the same extent in all age groups in recent the effects years. The of the regional amortisation differences in amortisation behaviour are generally requirement small, are although analysed amortisation is slightly less common in the metropolitan on the 2014 areas sample. than FI in has the now based 0 0 0 0-25 25-50 50-70 70-85 above 85 18-30 31-50 51-65 65+ Loan-to-value ratio, per cent rest of Sweden. performed a similar analysis using Age 14 the 2015 data. An overview of the 2011 2012 2013 2014 2015 Debt-to-income ratio, per cent 2011 effects 2012 of 2013 the requirement 2014 2015 is shown in 2011 2012 2013 2014 2015 12 SWEDISH MORTGAGE HOLDERS 14. AMORTISATION AS A SHARE Effects of FI s proposed amortisation requirement In terms of 13 the share of households tha OF THE LOAN AMOUNT FOR 15. AMORTISING HOUSEHOLDS following the amortisation requiremen DIFFERENT DEBT-TO-INCOME FOR VARIOUS In December AGE GROUPS 2015, FI submitted a proposal for an am- 100 R1. SHARE OF AMORTISING RATIOS (Per cent) (Share of households, per cent) HOUSEHOLDS to-value ratios (Per cent) above 70 per cent. Of th 5 Daniel L. Greenwald (MIT Sloan) 100 Household amortise today (Diagram R2). Howeve ortisation Debt and requirement Monetary Policy for con- January 7, 2016 7 / 8 Source: FI s sample. 1 0-25 25-50 50-70 70-85 above 85 0-150 150-300 Loan-to-value ratio, per cent 2011 2012 2013 2014 2015 300-450 450-600 600-750 750-900 above 900 Source: FI s sample. Source: FI s sample. 2 1 0 0-150 Younger households amortise to a gre (Diagram 15). This is probably mainly loan-to-value ratios. However, amortis less the same extent in all age groups i ferences in amortisation behaviour are tisation is slightly less common in the rest of Sweden. 150-300 300-450 450-600 600-750 750-900 Debt-to-income ratio, per cent above 900 Source: FI s sample. Source: FI s sample.
Conclusion Nice empirical exercise estimating consumption response to mortgage interest rate change. - But cash flow channel may not yet be fully isolated. Possible fix: - Identify households that are corner vs. interior on principal payments. - Challenge: avoid selection issues. Otherwise, just need to be careful interpreting results. - Perfect to analyze effect of mortgage indexation on monetary policy. - Trickier for other policies (e.g., debt forgiveness, maturity extension). Silver lining: restricting sample to households with interior solution could yield great estimates of EIS. - Isolates intertemporal substitution channel from other GE effects. Daniel L. Greenwald (MIT Sloan) Household Debt and Monetary Policy January 7, 2016 8 / 8