Credit risk model for SME in the Netherlands

Original by J.Hessel Veurink, M. Mastrogiacomo, 2014, 47 pagesHamster_gagarin_linkedin
hamster writter This summary note was posted on 11 January 2017, by in Credit risk Finance #, #
  • The data is far from ideal
  • Concludes that SMEs are less dependent on the macroeconomic environment than corporates
  • Default multipliers for stress test sensitivity tests for SMEs are usually proxies based on international evidence.
  • Missed information on interest rate so had to model it
  • 2 types of SME models: Structural (year reports, shares) and reduced form (balance sheet information, macroeconomic variables)
  • Schleifer and Vishny (1992) suggested that the selling asset prices of firms in financial distress depend on the financial state of their industry peers.
  • Most risk held by Irish loans was concentrated in the speculative sectors: Construction, Real Estate and Hotel& Restaurants in 2010.
  • Fidrmuc and Hainz (2010) find that in sloviakian SME loans between 200 and 2005, industry effect influences the PD
  • Five different standard categories: profitability, liquidity, solvency, activity, leverage ratios that is 1) working capital to total assets ratio, 2) retained earnings to total assets ratio, 3) earnings before interest, taxes and amortization (EBITDA) to total asset ratio, 4) market value of equity to book value of total debt ratio, 5) sales to total assets ratio.
  • Lennox (1999) claims that well specidfied logit and probit models outperforms MDA
  • Altman and Sabato (2007) apply logit on US SMEs between 1994 and 2007.
  • Diestch and Petey (2002) on German and French SME find that SMEs are generally less dependent on the state of the economy than larger firms but Jacobson et al (2005) conclude that SME are more dependent on the macroeconomic conditions
  • Simon and Rolwes (2009) in the Netherlands, show that there is a strong negative effect of GDP growth on the PD
  • Wilson (1997) performs stress test on the GDB growth. 3 steps: 1) PD explained by a set of macroeconomic variables, 2) macroeconomic variables independently follow an autoregressive model in order to build macroeconomic time series, 3) parameters are estimated and the future path of the macroeconomic variables is simulated
  • This model considers defaults independent of one another. This means for instance that the model treats the unemployment rate and GDP as unrelated variables. Consequently, scholar applied vector autoregression models (VAR) to take into account the interdependencies among the multiples macroeconomic time series.
  • Stress test methodologies survey by Foglia (2009)
  • Eklund (2001) use bank sheet variables, financial ratios and industry variables but also simulated next year’s balance sheets and income statements using macroeconomic projections instead of using last year’s data. Uses a discrete hazard model to obtain the probability of default
  • Annual SME default rate of NL is ~2.2% from 2005-2012
  • Amount of the loan influences default rate
  • Financial health based on accounting ratios is the main indicator of default in default probabilities studies for SMEs
  • Shumway (2001) shows that estimators of the static logit are positively biased and inconsistent. Secondly they cannot incorporate time variant variables such as macroeconomic variables. Suggests suing discrete hazard model.
  • Discrete hazard model is a specific form of survival analysis
  • Shumway (2001= explicitly shows that in discrete time the hazard model is equivalent to a multi-period logit model
  • Becker et al (1998) use dummies as baseline for years
  • Most default models make use a macroeconomic variable as baseline instead of the dummies above.
  • Shumway suggests using the natural log of the firm’s age however this baseline hazard is independent of the economic conditions.
  • Hillegeist (2004) use the economic wide bankruptcy rate of the previous year as a proxy for the macroeconomic environment
  • Nam et al (2008) use the volatility of the foreign exchange rate as baseline without considering dummies
  • Find that in contrast with Hillegeist dummies does have a positive impact on the model
  • Have to take care of left censoring, and use age of the loan to mitigate this problem
  • Use GDP growth (-)m interest rate (+), Age of loan (-), Ebitda (-) and turn over (+). Brackets show expected effect.
  • Duffie et all (2007) point out that extrapolation of models for one year ahead default probabilities to longer time horizon fails if one does not incorporate the long term dynamics of the underlying firm specific covariates
  • Increasing borrowing costs leads to more difficulties in meeting financial obligations (interest rate)
  • High loan age before entering the sample significantly decreases the probability of default.
  • Area under the ROC curve is 0.6696.
  • McCann and McIndoe-­‐Calder (2012). They find an area under the ROC curve of 0.7749. Their model is based on a data set with full information of the balance sheet of the SME,
  • Dietsh and Petey (2002) find that SME are less dependent on the state of the economy