Banque nationale de Belgique SA · EURONEXT:BNB
Banque nationale de Belgique SA — Investor Presentation (June 2016)
40 slides from Banque nationale de Belgique SA's investor presentation, Jun 20, 2016
Slide 1: Credit and Crises
Credit and Crises June 20, 2016 c 2016 by Christopher A. Sims. This document is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License. http://creativecommons.org/licenses/by-nc-sa/3.0/
Slide 2: Aim of the talk
Aim of the talk Heading off the emergence of a spurious “stylized fact”. 1
Slide 3: Aim of the talk
Aim of the talk Heading off the emergence of a spurious “stylized fact”. Arguing for careful econometrics in studying the relation between financial variables and the macroeconomy, taking an approach that is • multiple-equation, • Bayesian, • and weakly structural. 1
Slide 4: Why multiple equation?: Positive links between credit and growth
Why multiple equation?: Positive links between credit and growth • “Financial depth” is treated as a predictor of growth in development regressions. 2
Slide 5: Why multiple equation?: Positive links between credit and growth
Why multiple equation?: Positive links between credit and growth • “Financial depth” is treated as a predictor of growth in development regressions. • Over long spans of time, credit grows relative to gdp as gdp increases. 2
Slide 6: Why multiple equation?: Positive links between credit and growth
Why multiple equation?: Positive links between credit and growth • “Financial depth” is treated as a predictor of growth in development regressions. • Over long spans of time, credit grows relative to gdp as gdp increases. • Countries with high gdp per capita tend to have high credit/gdp. 2
Slide 7: Negative links between credit and growth
Negative links between credit and growth • Standard loan contracts are simple, easily verifiable, when the borrower can repay. 3
Slide 8: Negative links between credit and growth
Negative links between credit and growth • Standard loan contracts are simple, easily verifiable, when the borrower can repay. • When the borrower can’t repay, they throw ownership of assets into an uncertain state, to be resolved by an expensive and slow process, which hurts economic efficiency. 3
Slide 9: Negative links between credit and growth
Negative links between credit and growth • Standard loan contracts are simple, easily verifiable, when the borrower can repay. • When the borrower can’t repay, they throw ownership of assets into an uncertain state, to be resolved by an expensive and slow process, which hurts economic efficiency. •
Slide 10: How feedbacks affect policy issues
How feedbacks affect policy issues • Credit is not in itself bad, clearly, so policy that simply inhibits credit growth is a mistake. 4
Slide 11: How feedbacks affect policy issues
How feedbacks affect policy issues • Credit is not in itself bad, clearly, so policy that simply inhibits credit growth is a mistake. • If policy-induced interest rate increases do inhibit credit growth (which seems likely), this does not mean interest rate increases are a good idea once it is clear
Slide 12: How feedbacks affect policy issues
How feedbacks affect policy issues • Credit is not in itself bad, clearly, so policy that simply inhibits credit growth is a mistake. • If policy-induced interest rate increases do inhibit credit growth (which seems likely), this does not mean interest rate increases are a good idea once it is clear
Slide 13: How feedbacks affect policy issues
How feedbacks affect policy issues • Credit is not in itself bad, clearly, so policy that simply inhibits credit growth is a mistake. • If policy-induced interest rate increases do inhibit credit growth (which seems likely), this does not mean interest rate increases are a good idea once it is clear
Slide 14: Empirical negative links between credit and growth
Empirical negative links between credit and growth 1. Schularick and Taylor AER 2. Jorda-Schularick-Taylor 3. Mian-Sufi-Werner 4. BIS work on this 5
Slide 15: Some reasons I'm not fully convinced
Some reasons I’m not fully convinced • Except for Mian-Sufi-Werner, this work filters outcomes via a “crisis” classification scheme. • They rely mainly on single-equation methods, mostly with a single “forecast horizon”. • 3 also use a panel VAR, but only with 3 variables — two credit quantity varia
Slide 16: The need for a multiple equation approach
The need for a multiple equation approach • Because the relation between credit and real activity is likely to have a different sign, according to what frequency we look at or to what source of disturbance to the economy we examine, any analysis useful for policy must take a multi-equation approach
Slide 17: The need for a multiple equation approach
The need for a multiple equation approach • Because the relation between credit and real activity is likely to have a different sign, according to what frequency we look at or to what source of disturbance to the economy we examine, any analysis useful for policy must take a multi-equation approach
Slide 18: The need for a multiple equation approach
The need for a multiple equation approach • Because the relation between credit and real activity is likely to have a different sign, according to what frequency we look at or to what source of disturbance to the economy we examine, any analysis useful for policy must take a multi-equation approach
Slide 19: Why “semi-structural”
Why “semi-structural” • DSGE’s are story-telling devices, not science. • They’re better than RBC or old Keynesian models, because they are formulated as explicit multivariate time series models that can be compared in fit to structural VAR’s. • SVAR’s came first: DSGE’s were built to match SVAR resu
Slide 20: DSGE limitations
DSGE limitations • But they don’t fit as well as SVAR’s. • We don’t really believe the stories they tell. (e.g., K, I, C, L, W and P are fictions) • They are awkward tools when, as now, we are uncertain as to whether and how we should be expanding the list of variables that enter our macro models. 9
Slide 21: A puzzle: Mian-Sufi-Verner results
A puzzle: Mian-Sufi-Verner results 10
Slide 22: Impulse responses with logs of rgdp, aha, ana, no Turkey
pr[ir, ic, , center] pr[ir, ic, , center] pr[ir, ic, , center] 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.00 0.02 0.04 0.06 0.08 −0.005 0.000 0.005 0.010 0.015 0.020 0.025 0 10 20 Y1 30 40 0 10 20 Y2 30 40 irf's with logs of rgdp, aha, ana, no Turkey 0 10 20 Y3 30 40 11
Slide 23: Equations
Equations A(L)yi(t) = ci + Λ−1 i εi (t) yi(t) n × 1; εi(t) ∼ N (0, I), i.i.d.; Λi diagonal; sum of diagonal elements of Λ over i is a vector of 1’s. A−1(L)Λi represents impulse responses for country i, while A−1(L) by itself is a kind of harmonic average impulse response. 12
Slide 24: Identification through heteroskedasticity
Identification through heteroskedasticity Under the (strong) assumption that all differences across countries are captured in the ci and Λi parameters, so long as there are even as many as two countries across which the ratios of the diagonals of Λi are all different, the system — and hence the resp
Slide 25: Identification through heteroskedasticity
Identification through heteroskedasticity • This is not “identification by sign restrictions on impulse responses”. That does not produce exact identification, even in large samples. 14
Slide 26: Identification through heteroskedasticity
Identification through heteroskedasticity • This is not “identification by sign restrictions on impulse responses”. That does not produce exact identification, even in large samples. • Of course this may be too good to be true. It remains to be seen how well it works in practice. 14
Slide 27: Identification proof
Identification proof Σ1 = A−1Λ1(A0)−1 , Σ2 = A−1Λ2(A0)−1 ∴ Σ−1 0 −1 0 −1 1 Σ2 = A Λ1 Λ2 (A ) This last matrix has the columns of A0 as eigenvectors and the diagonal of Λ−1 −1 1 Λ2 as eigenvalues. As long as the diagonal elements of Λ1 Λ2 are all distinct, the columns of A0 (rows of A) are uniquely d
Slide 28: CAS semi-replication with prior, quarterly data, ID through heteroskedasticity
CAS semi-replication with prior, quarterly data, ID through heteroskedasticity 16
Slide 29: Modal responses to shocks, 30 countries
ana aha defl rgdp 0.000 0.005 0.010 0.015 0.020 0.025 0.030 −0.02 −0.01 0.00 0.01 0.02 0.03 0.04 0.05 0.000 0.005 0.010 0.015 0.020 −0.005 0.000 0.005 0.010 0 10 1 20 30 40 0 10 2 20 30 40 0 10 3 20 30 Modal responses to shocks, 30 countries 40 0 10 4 20 30 40 17
Slide 30: Remarks on the replication
Remarks on the replication • The variables are real gdp and nominal credit to households and to firms, all logged. • A positive and persistent shock to prices would increase the nominal credit variables, generate a restrictive monetary policy response, and thereby dampen GDP growth. • In MSV’s origi
Slide 31: A bigger model
A bigger model • A panel data structural VAR for 13 countries. • Country-specific constant terms, but the rest of the VAR constant across countries. • Identification through heteroskedasticity, with the variances of structural shocks allowed to vary across countries. • Probably most important: An in
Slide 32: Data
Data 13 countries: Austria, Belgium, Canada, Denmark, France, Italy, Japan, Netherlands, Norway, Sweden, United Kingdom, US, and Switzerland. Sample periods differ across countries, based on data availability. 6 variables: real property prices (rp), real gdp (rgdp), short interest rate (usually 1-mo
Slide 33: Estimation methods
Estimation methods • 108 free parameters in A0, Λ, 180 more in A+(L). • A Bayesian approach is essential. A pure dummy-observation Minnesota prior (including pulling toward unit roots and toward cointegration) was used conditional on A0, Λ. • A0 elements had mean 100 on the diagonal, 0 off diagonal,
Slide 34: Results
credit defl pc r rgdp rp −0.04 −0.02 0.00 0.02 −0.015 −0.010 −0.005 0.000 0.005 0.010 −0.20 −0.15 −0.10 −0.05 0.00 0.05 0.10 −0.001 0.001 0.002 0.003 0.004 0.005 −0.005 0.000 0.005 0.010 −0.03 −0.02 −0.01 0.00 0.01 0.02 0.03 0.04 0 10 1 20 30 40 0 10 2 20 30 40 0 10 3 20 30 40 0 Results 10 4 20 Aver
Slide 35: Worth noting
Worth noting • The identification approach seems to have thrown up one shock that behaves like a usual SVAR monetary policy shock (number 3) • There is no shock that produces an increase in credit followed by below- trend gdp. • There is one shock (number 4) that produces steady exapansion of credit
Slide 36: Is it the credit expansion or the source of the inflation that dampens output growth?
which is followed by an interest rate rise. Is it the credit expansion or the source of the inflation that dampens output growth? 24
Slide 37: Why so different?
Why so different? • Including interest rates and monetary policy reactions is important. Possibly the timing studies are just picking up the effects of monetary policy reacting to inflation and credit growth. • No accounting for cross-country shock correlations in the same year, and these are high.
Slide 38: Unit roots / initial conditions
• Unit roots / initial conditions. The strongest priors toward unit roots were on the biggest model, and this should thus have the least problem of this sort. But requires further exploration. 26
Slide 39: Why so different?
Why so different? • Country lists and data spans don’t match. • Symmetric models don’t make sense: US r should probably enter everywhere else. This might mitigate cross-country correlation of shocks, which is hard to interpret in an SVAR. • The 6-variable model shown here includes only the household
Slide 40: Conclusion
Conclusion • We should not be treating the notion that rapid credit expansion implies future growth slowdowns as an established stylized fact. • Using weakly identified models (mainly SVAR’s) that allow testing of assertions about which credit indicators are statistically reliable is important. • In
Banque Nationale De Belgique — Investor Presentation (June 2016) on 6ix