8107774987_5de2bd7d2d_b.jpg
VoxEU Column Monetary Policy

Re-evaluating the result that the costs of ‘leaning against the wind’ exceed the benefits

The IMF and the Federal Open Market Committee have both suggested that the costs of ‘leaning against the wind’ exceed the benefits. This column responds to claims that the results of the author's research backing up this conclusion could be overturned. It argues that the alternative assumptions necessary to overturn the result are unrealistic, and that the finding that the costs of the policy exceed the benefits therefore seems to be robust.

‘Leaning against the wind’ refers to conducting, for financial-stability purposes, a tighter monetary policy (i.e. setting a higher policy interest rate) than would be justified by standard flexible inflation targeting if policymakers disregarded the possibility of a financial crisis. It has been promoted by the Bank for International Settlements (BIS 2014). Regarding the costs and benefits of leaning against the wind, an IMF Staff Report concluded that, “based on current knowledge, the case for leaning against the wind is limited, as in most circumstances costs outweigh benefits” (IMF 2015). The Federal Open Market Committee (FOMC) reported from a discussion on the same topic that:

“Most participants judged that the benefits of using monetary policy to address threats to financial stability would typically be outweighed by the costs associated with deviations from the Committee’s employment and price-stability objectives induced by such actions.” (FOMC 2016)

If we are to make a cost-benefit analysis of leaning against the wind, we need numerical estimates of the costs and benefits. In Svensson (2017a) and previous versions, I provided a simple method combining existing empirical estimates into numerical estimates. The costs of leaning against the wind are higher unemployment and lower inflation both if a crisis is avoided and, importantly, if a financial crisis were to occur. Benefits include a lower probability of, or magnitude, crisis.

The analysis shows that the costs substantially exceed the benefits. This is because current empirical estimates of the policy-rate effects on the probability and magnitude of a crisis are so small as to prevent benefits from becoming larger than costs. The robustness of this result is supported by a detailed and extensive sensitivity analysis in the main text, and the appendix, of Svensson (2017a). I have also summarised the method and its result in a previous Vox column (Svensson 2016a).

A paper by Adrian and Liang (2016a) questioned the robustness of the result. Instead of using the sensitivity analysis provided in Svensson (2017a), the authors conducted an alternative sensitivity analysis using a simple Excel spreadsheet example that I used in the slides presenting  a previous paper of mine (Svensson 2016b) at a conference.1 This was a simplified two-period example to illustrate the mechanisms and orders of magnitude of the relevant effects, but its simplifications and short-cuts would make it unsuitable for a serious, reliable sensitivity analysis. For that, my original framework in Svensson (2017a) is more appropriate, given that it uses empirically estimated costs and benefits with empirical lags and dynamics over a 40-quarter period.

In a Vox column summarising their paper, Adrian and Liang stated that:

“The conclusion that costs exceed benefits is very sensitive to reasonable alternative assumptions about three key parameters: the severity of a crisis; the probability of a crisis; and the sensitivity of the probability of a crisis to monetary policy.” (Adrian and Liang 2016b, italics added)

In a new paper (Svensson 2017b), I have provided some additional sensitivity analysis of my previous result, and examine in some detail the alternative assumptions that would have been required to overturn it. I conclude that these alternative assumptions would hardly be reasonable or realistic. They involve large deviations from existing empirical estimates in the literature. The alternative assumptions suggested by Adrian and Liang imply an effect of debt on the magnitude of a crisis that is more than 40 standard errors larger than the estimate of Flodén (2014) and more than 11 standard errors larger than an estimate that follows from Jorda et al. (2013). They also imply an effect of credit on the probability of a crisis that is more than 13 standard errors larger than the estimate of Schularick and Taylor (2012).

The issues raised by Adrian and Liang concern the policy-rate effect on the magnitude of a crisis, the probability of a crisis starting, the policy-rate effect on the probability of a crisis, and the role of the risk-taking channel in the cost-benefit analysis. I discuss these in turn.

The policy-rate effect on the magnitude of a crisis

In the framework presented in Svensson (2017a), the main cost of leaning against the wind was an increase in the unemployment rate, not only if no financial crisis occurs (a ‘non-crisis’) but also, importantly, if a financial crisis occurs (a ‘crisis’).2 Assume that a policy rate that is 1 percentage point higher during four quarters increases the non-crisis unemployment gap in between six and eight quarters from 0 to 0.5 percentage point in a non-crisis. This is in line with the standard estimate of the Sveriges Riksbank of the policy-rate effect on unemployment. Furthermore, assume at first that the magnitude of a crisis, net of any policy response during the crisis, is given and represented by a benchmark increase in the unemployment rate of 5 percentage points (assumption from Sveriges Riksbank 2013). The higher policy rate increases the crisis unemployment gap by 0.5 percentage point, from 5 to 5.5 percentage points.

The cost of a higher unemployment gap in a crisis is the main component of the cost of leaning against the wind. But if leaning against the wind reduces the magnitude of a crisis, this will in turn reduce the net cost of leaning against the wind. As Adrian and Liang (2016a) noted, the alternative assumption that the higher policy rate would reduce the crisis increase in the unemployment rate by 0.5 percentage points, from 5 to 4.5 percentage points, would overturn the result that that the costs of leaning against the wind exceed the benefits and make the costs and benefits equal. A larger reduction in the crisis increase in the unemployment rate would then make the costs less than the benefits.

The issue, then, is whether leaning against the wind could realistically cause such a large reduction in the magnitude of a crisis. To assess its effect on the magnitude of a crisis, I combined the Sveriges Riksbank (2014) estimate of the policy-rate effect on the debt-to-income ratio with Flodén’s (2014) estimate of the effect of the debt-to-income ratio on the increase in the unemployment rate in the OECD countries during the Great Recession. I found that Flodén’s estimate was so small that the policy-rate effect on the magnitude could be disregarded. Furthermore, counter to Adrian and Liang (2016a), the estimates of Jorda et al. (2013) of the effect of the debt-to-GDP ratio on the fall in GDP during a financial crisis was similar to Flodén’s estimate. This was also the case for the estimate of Krishnamurthy and Muir (2016).3

I have explicitly calculated how much stronger the effect of the debt-to-income ratio on the crisis increase in the unemployment rate must be for break-even, that is, for the costs and benefits of leaning against the wind to be equal. The required effect must be 17 times as large as Flodén’s estimate. This required effect is 40 standard errors larger than Flodén’s estimate, and 11 standard errors larger than the estimate that follows from Jorda et al. (2013). In this unrealistic case, leaning against the wind would indeed have the benefit of making the crisis increase in the unemployment rate about 0.5 percentage points smaller. This is what would be required for break-even. For the benefits of leaning against the wind to exceed the costs, an even larger deviation would be required.

The probability of a crisis

Regarding the probability of a crisis, Adrian and Liang (2016a) suggested that a higher probability of a crisis would imply that a smaller reduction of the magnitude of a crisis is sufficient to overturn the result that costs exceed benefits. This is not true in the framework in Svensson (2017a). For a higher probability of a crisis, a 40 standard errors larger estimate than Flodén’s (and about 11 standard errors larger than the estimate that follows from Jorda et al. 2013) would still be required for break-even. This is because for break-even, the policy-rate effect on the magnitude of a crisis still has to offset the policy-rate effect on the non-crisis unemployment rate, regardless of the probability of a crisis.

The policy-rate effect on the probability of a crisis

Regarding the policy-rate effect on the probability of a crisis, my estimate was constructed by combining the Sveriges Riksbank (2014) estimate of the policy-rate effect on real debt (and thereby on real debt growth) and Schularick and Taylor’s (2012) estimate of the effect of real credit growth on the probability of a crisis. This implies a rather small empirical policy-rate effect on the probability of a crisis. Adrian and Liang suggested that I underestimated this effect. They argue that an alternative assumption would be that a 1 percentage point increase in the policy rate would reduce the probability of a crisis by as much as 1 percentage point.

To further examine this issue, I calculated how much stronger an effect would be needed for break-even. I stacked the cards in favour of leaning against the wind by assuming monetary non-neutrality and a permanent effect on real debt from a temporary policy-rate increase. Then the required effect would be about six times as large as Schularick and Taylor’s (2012) estimate. This would be an effect that is 13 standard errors larger than their estimate. The assumption of such a large effect implies that a 1 percentage point higher policy rate would lead to maximum fall in the probability of a crisis of about 1.3%, similar to the alternative assumption of Adrian and Liang. This is clearly an assumption that deviates too much from existing empirical estimates to be realistic.

The risk-taking channel

Regarding the risk-taking channel, the estimates of the Sveriges Riksbank (2014) and Schularick and Taylor (2012) were reduced-form estimates and may have picked up any effects through the risk-taking channel. Adrian and Liang (2016a) suggested that the risk-taking channel would increase the policy-rate effect on the magnitude and probability of a crisis.

I am not aware on any empirical support for a risk-taking channel large enough to matter in this context. A thorough analysis in Dell’Ariccia et al. (2013) of the effect of the federal funds rate on the riskiness of bank loans in the US found that the risk-taking channel was statistically significant but economically insignificant. This is because the effect of a 1 percentage point increase in the federal funds rate is only 3.6% of the standard deviation of the risk-taking measure. Thus, the risk-taking channel is insignificant compared to the effect of other factors on the risk-taking measure, and therefore unlikely to matter for the estimate of the policy-rate effect on the probability of a crisis.

Given this, and the extensive sensitivity analysis already done in the current and previous versions of Svensson (2017a), my conclusion is that, so far, the result that the costs of leaning against the wind exceed the benefits by a substantial margin stands up to scrutiny. Given current knowledge and existing empirical estimates, the result seems to be robust and not sensitive to reasonable alternative assumptions.

References

Adrian, T and N Liang (2016a), “Monetary Policy, Financial Conditions, and Financial Stability,” Federal Reserve Bank of New York Staff Report No. 690, revised December 2016.

Adrian, T and N Liang (2016b), “Monetary Policy, Financial Conditions, and Financial Stability,” VoxEU, 14 August. 

Bank for International Settlements (BIS) (2014), 84th Annual Report.

Dell’Ariccia, G, L Laeven, and G Suarez (2013), “Bank Leverage and Monetary Policy’s Risk-Taking Channel: Evidence from the United States,” IMF Working Paper WP/13/143, forthcoming in Journal of Finance.

Federal Open Market Committee (FOMC) (2016), “Minutes of the Federal Open Market Committee, April 26-27, 2016”, Federal Reserve Board.

Flodén, M (2014), “Did Household Debt Matter in the Great Recession?” Supplement to blog post on Ekonomistas.se, Sveriges Riksbank.

IMF (2015), “Monetary Policy and Financial Stability,” Staff Report, International Monetary Fund, www.imf.org.

Jorda, O, M Schularick, and S M Taylor (2013), “When Credit Bites Back,” Journal of Money, Credit and Banking 45, Supplement(2), 3-28.

Krishnamurthy, A, and T Muir (2016), “How Credit Cycles across a Financial Crises,” Working Paper.

Schularick, M, and A M Taylor (2012), “Credit Booms Gone Bust: Monetary Policy, Leverage Cycles, and Financial Crises, 1870-2008,” American Economic Review 102, 1029-1061.

Svensson, L E O (2016a), “Cost-Benefit Analysis of Leaning Against the Wind: Are Costs Larger Also with Less Effective Macroprudential Policy?” VoxEU, 12 January. 

Svensson, L E O (2016b), “Monetary Policy and Macroprudential Policy: Different and Separate,” forthcoming in Canadian Journal of Economics.

Svensson, L E O (2017a), “Cost-Benefit Analysis of Leaning Against the Wind: Are Costs Larger Also with Less Effective Macroprudential Policy?” CEPR Discussion Paper DP11739, revision of IMF Working Paper WP/16/03.

Svensson, L E O (2017b), “How Robust Is the Result That the Cost of ‘Leaning Against the Wind’ Exceeds the Benefit? Response to Adrian and Liang,” CEPR Discussion Paper DP11744.

Sveriges Riksbank (2013), “Financial Imbalances in the Monetary Policy Assessment,” Monetary Policy Report, July.

Sveriges Riksbank (2014), “The Effects of Monetary Policy on Household Debt,” Monetary Policy Report, February.

Endnotes

[1] The simple excel sheet is available here.

[2] More precisely, the loss function used is an indirect loss function of the deviation of the actual unemployment rate from the unemployment rate that would be optimal under flexible inflation targeting. This way the loss function incorporates not only the loss from unemployment deviating from its long-run sustainable rate, but also the loss of inflation deviating from the inflation target.

[3] Adrian and Liang refer to Jorda et al. (2013) as implying a larger effect – but they do not translate the result of Jorda et al. (2013)  into a form comparable to that of Flodén (2014).

2,625 Reads