Session on Macro Risk. Discussion. Olivier Loisel. crest. 8 th Financial Risks International Forum Scenarios, Stress, and Forecasts in Finance

Transcription

1 Session on Macro Risk Discussion Olivier Loisel crest 8 th Financial Risks International Forum Scenarios, Stress, and Forecasts in Finance Paris, March 31, 2015 Olivier Loisel, Crest Discussion on Macro Risk Paris, March 31, / 10

2 Two papers on macro risk Two nice papers, with two different perspectives on macro risk : 1 Diris-Keijsers-Kole: correlation between the business cycle and the default rate on bank loans the loss given default on bank loans 2 : crisis-predicting properties of country-level macro, financial, and banking indicators within-country and cross-country financial linkages Olivier Loisel, Crest Discussion on Macro Risk Paris, March 31, / 10

3 Data and model Data sources: bank-loan variables: Pan-European Credit Database Consortium macro variables: OECD Data content: bank-loan variables: DR, LGD, seniority, security, asset class, industry macro variables: GDP, IP, UR Model overview: LGD: mixture of two normal distributions (good and bad losses) DR: Bernoulli distribution The default and bad-loss probabilities depend on the loan characteristics the same latent factor α t The model is estimated with the Expectation Maximization algorithm Olivier Loisel, Crest Discussion on Macro Risk Paris, March 31, / 10

4 Main results 1 DR and LGD respond to α t in the same direction 2 α t is related to the business cycle: negatively to GDP and IP (in the same quarter) positively to UR (three or four quarters later) 3 Observable bank-loan characteristics matter 4 Most of the variation is due to changes in the probability of a bad loss Olivier Loisel, Crest Discussion on Macro Risk Paris, March 31, / 10

5 Comments I: data Are the OECD and PECDC datasets matched at the European level? If so, is there evidence that all or most borrowers are European? And that borrowers nationality is representative of Europe? Yearly bank-loan observations are affected to Q3 (other Qs missing): is the third quarter representative of the year (e.g., in 2008)? why not instead transform quarterly data into yearly data? Why treat non-default observations as missing values? Why not consider information on the maturity of loans? the period at which default occurs over the loan lifespan? What about distinguishing crisis from non-crisis periods? Olivier Loisel, Crest Discussion on Macro Risk Paris, March 31, / 10

6 Comments II: implications Implications for banks: should they conduct a similar analysis using more data (price, quality, clauses, etc) on their own loans? How would the results be affected? Implications for micro-/macro-prudential authorities (Basel III): should they combine credit data with macro forecasts? But then what about the Lucas critique? Olivier Loisel, Crest Discussion on Macro Risk Paris, March 31, / 10

7 Data and measure of banking-sector centrality Data sources: banking crises: ESCB macro, financial, banking indicators: ECB-BSI within-country financial linkages: ECB-EAA cross-country financial linkages: ECB-BSI Measure of banking-sector centrality: four candidate measures: degree-in, degree-out, betweenness, closeness four candidate instruments: loans, deposits, securities, shares principal-component analysis (PCA) on the 4 4 = 16 variables Olivier Loisel, Crest Discussion on Macro Risk Paris, March 31, / 10

8 Model and main results Main features of the model: logit analysis threshold values for probabilities goodness-of-fit measure called usefulness Main results: considering cross-border linkages increases usefulness considering all linkages increases usefulness further loans and securities are the most important instruments for usefulness Robustness analysis with respect to definition of usefulness forecast horizon probability thresholds real-time analysis Olivier Loisel, Crest Discussion on Macro Risk Paris, March 31, / 10

9 Comments I: data What is the definition of a banking crisis? Data over 2000Q1-2012Q1 for 14 European countries: how many banking-crisis observations all in all? how many over 2000Q1-2005Q2 (in-sample analysis)? how correlated are they over time and across countries? What are the main estimated PCA components? Olivier Loisel, Crest Discussion on Macro Risk Paris, March 31, / 10

10 Comments II: cross-border transmission of crises Could networks affect both crisis occurrence and crisis severity? Why not mix national indicators with cross-border linkages? use national indicators to compute domestic-crisis probabilities use cross-border linkages to infer domestic- or imported-crisis prob. Why not consider the joint distribution of national banking crises? What implications for the ESRB and national prudential authorities? Olivier Loisel, Crest Discussion on Macro Risk Paris, March 31, / 10