Credit performance of the UK SMEs Through the Crisis Jake Ansell Credit Research Centre, The University of Edinburgh Business School J.Ansell@ed.ac.uk Joint work with Dr Galina Andreeva, Paul Orton, Dr Ma Yigui and Ma Meng 1 Outline • Background • Data • Cross-sectional Analysis • Panel Data with Dummies • Panel Data with Macroeconomic Variables • Future plans? • Conclusion 2 SMEs - Cornerstone of the Economy Globally 95% Businesses are SMEs, 50% of economic value, 55% of all innovations EU 99% Businesses are SMEs, 68% of total employment, 63% of overall business turnover UK 99% Businesses are SMEs, 59% of total employment, 50% GDP Similar picture for Asian economies 3 Lending in UK • Concern over lending to SMEs in UK (£991m in 2008, £566m in 2010) • Prudent lending requires more stringent criterion • SMEs more conservative in recessionary periods • Anecdotal information that some SMEs feel credit constraints 4 Credit Scoring and SMEs • Business Managers assessing clients – picking winners (Very old model) • Business Relationship Management – plausible for high value clients less for SMEs • But need fast efficient methods credit decisions for many small businesses – Credit Scoring • More recently ‘Management Capability’ – Ma Yigui, Andreeva and Ansell (2011) 5 Credit risk approaches • Lending to individuals • Lending to businesses - Relatively small amounts of money lent to a large number of customers - focus more on prediction, less on causality - Management Science and Data Mining - Large amounts of money lent to a relatively small number of businesses - focus more on causality, less on prediction - Finance and Accounting 6 Data • • • • • There are about 5 million SMEs in UK Not all SMEs borrow from banks Database from a Credit Agency Over 2 million enterprises Recorded each April: 2007, 2008, 2009 & 2010 7 Data • Financial Impairment: Good/Bad • General Information: legal form, region, SIC, # Employees, Age of Company • Directors’ Information: # Directors, Ownership, Changes etc • Previous Credit history: DBT, judgements etc • Accounting Information: Common financial variables and financial ratios 8 Impairment Rate in UK (%) 18 16 14 12 10 8 Series1 6 4 2 0 2006 2007 2008 2009 2010 2011 9 Impairment Rate by Region 25 20 London Scotland North East 15 North West West Midlands Wales 10 South West East Midlands South East 5 0 2007 2008 2009 2010 10 Impairment by SIC code 11 Impairment by Age 12 Initial Analysis • • • • Cross-Sectional Analysis Logistic Model Predicting Default Model Used Weights of Evidence Stepwise Regression using % change in Cox & Snell (Nagelkerke) • Interest in Performance and Variable Inclusion 13 Cox and Snell/Nagelkerke 2007 2008 All 0.308 0.517 Start-Up 0.149 0.324 Start-Up 0.329 0.500 Non SU 0.052 0.196 Non SU 0.205 0.427 All 0.211 0.401 Start-Up 0.235 0.390 Start-Up 0.238 0.393 Non SU 0.126 0.336 Non SU 0.148 0.372 All 0.120 0.300 0.207 0.390 2009 2010 All 14 AUROC Results In Sample CI CI 2007 Difference 0.82 0.816 0.824 0.82 0 0.82 0.8155 0.8245 0.82 -0.003 Non SU 0.794 0.785 0.803 0.793 0.002 All 0.852 0.849 0.854 0.841 0.011 0.84 0.837 0.844 0.826 0.014 0.843 0.837 0.85 0.837 0.006 All Start2007 Up 2008 StartUp Non SU 15 AUROC Results In Sample CI CI 2007 Difference All 0.886 0.884 0.888 0.876 0.01 2009 Start-Up 0.868 0.865 0.87 0.853 0.015 Non SU 0.87 0.865 0.874 0.889 -0.019 All 0.851 0.849 0.854 0.84 0.011 2010 Start-Up 0.83 0.826 0.833 0.811 0.019 Non SU 0.85 0.845 0.856 0.851 -0.001 16 2Comments • Whilst R2 are low the predictive quality is high in sample and out sample • No out of time results • Modelling was naïve • There is some stability over variables or type of variables • There is stability over time – could be due to nature of variables employed 17 Panel Analysis • Obviously can trace behaviour of individual enterprises over time • But only have 4 observation points • Modelling default – No loss measurment • Good = 0, Bad = 1 • Logit Panel Data Model: Log(Pg/Pb) = ai+bixii+di+sii 18 Panel Analysis • Produce Cross-Section Models each Year • Using Panel Sample Tracking Enterprises • Panel Analysis and Panel Analysis with Dummy for Years • Coefficients of Model, Performance, Absolute Mean Square Error 19 Impairment in Panel Sample 30.00 25.00 20.00 15.00 10.00 5.00 0.00 2007 2008 non_startups 2009 startups 2010 whole sample 20 Non-Start-Ups: SIC Code Non-Start-up SMEs 'Bad' Rate: 1992 SIC Code 0.3 missing angriculture 0.25 manufacture constraction retail trade 0.2 Axis Title hotels and restaurants transport, storage 0.15 financial intermediation property manegment 0.1 computers R&D legal consult 0.05 other professional education, health and social 0 private households with employee APR07 APR08 APR09 APR10 21 Non-Start-Up by Region 0.3 London 0.25 Scotland East Midlands 0.2 West Midlands North West 0.15 North East Wales/South West 0.1 South West South East 0.05 Other 0 APR07 APR08 APR09 APR10 22 Variable Start-Up Model 1. Legal Form 8. Total Value Of Judgements In The Last 12 Months 2. Company is Subsidiary 9. Number Of Previous Searches (last 12m) 3. 1992 SIC Code 10. Time since last derogatory data item (months) 4. Region 11. Lateness Of Accounts 5. Proportion Of Current Directors To Previous Directors In The Last Year 6. Oldest Age Of Current Directors/Proprietors supplied (Years) 7. Number Of Directors Holding Shares 12. Time Since Last Annual Return 13. Total Assets 23 Start-Up Models’ Coefficient 15 10 2007 2008 5 2009 2010 0 0 2 4 6 8 Panel Panel + Year -5 -10 Variable in list order 24 Start-Up Models’ Coefficient 1.8 1.6 1.4 2007 1.2 2008 1 2009 0.8 2010 0.6 Panel 0.4 Panel + Year 0.2 0 7 8 9 10 11 Variable in list order 12 13 25 Non-Start-up Variables 1. Legal Form 9. Number Of Previous Searches (last 12m) 2. Parent Company – derog details 10. Time since last derogatory data item (months) 3. 1992 SIC Code 11. Lateness Of Accounts 4. Region 12. Time Since Last Annual Return 5. No. Of ‘Current’ Directors 13. Total Fixed Assets As A Percentage Of Total Assets 6. Proportion Of Current Directors To Previous Directors In The Last Year 14. Debt Gearing (%) 7. PP Worst (Company DBT - Industry DBT) In The Last 12 Months 15. Percentage Change In Shareholders Funds 8. Total Value Of Judgements In The Last 12 Months 16. Percentage Change In Total Assets 26 Non-Start-up Results 1 0 0 2 4 6 -1 8 10 2007 2008 2009 -2 2010 Panel -3 Panel+Year -4 -5 Variable list order 27 Non-Start-up Results 1.4 1.2 1 2007 0.8 2008 0.6 2009 0.4 2010 0.2 Panel 0 -0.2 8 9 10 11 12 13 14 15 16 17 Panel+Year -0.4 -0.6 Incept + variable in listed order 28 Dummy Effects 0 1 2 3 -0.5 -1 -1.5 -2 non st -2.5 -3 -3.5 -4 29 Panel with Macro-economic Variable Currently Exploring of Macro-economic Variables: 1. 2. 3. 4. 5. 6. 7. UNEMPLOYMENT RATE INFLATION ANNUAL CHANGE CPI CPI ANNUAL CHANGE FTSE ALL SHARE INDEX CHANGE FTSE100 ANNUAL INDEX CHANGE FTSE 100 ANNUAL RETURN 30 Annual Macro variables Annual MVs GDP growth rate 40.0 ftsall index change rate unemployment 30.0 inflation 20.0 FTS100 change rate Axis Title 10.0 CPI rate 0.0 2005 -10.0 -20.0 -30.0 2006 2007 2008 2009 2010 2011 non_year dummy non_default rate st_year dummy st_default rate whole sample default rate -40.0 31 Averaged Annual Macro Variables Averaged Annual MVs 30.0 gdp_growth rate 25.0 cpir ftsall index 20.0 unemployment Axis Title 15.0 inflation FTS100 10.0 non_year dummy 5.0 non_default rate 0.0 2007 -5.0 2008 2009 2010 st_year dummy st_default rate whole sample default rate -10.0 32 Correlations gdp3 gdp3 FAI une infl F100 cpir FAI une infl F100 cpir 1 0.993632 1 0.791506 0.786689 1 -0.98125 -0.95905 -0.7189 1 0.978212 0.986059 0.781223 -0.9262 1 0.948904 0.972196 0.826953 -0.87191 0.982503 1 33 Start-Up Models 1 2 GDP Growth GDP Growth Lag 1 RPI RPI Lag 1 FTSE 100 FTSE 100 Lag1 3 GDP Growth Average last 3 Years 4 5 GDP Growth GDP Growth Lag 1 6 GDP Growth Average last 3 Years RPI Average Last 3 Years FTSE Average Last 3 Years 34 Start-up Models 2.0000 1.0000 0.0000 0 2 4 6 8 10 12 14 -1.0000 Series1 -2.0000 Series2 Series3 -3.0000 Series4 Series5 -4.0000 Series6 -5.0000 -6.0000 -7.0000 -8.0000 Incept + variable in listed order 35 Non-Start-Up Models 1 GDP Growth Average Last 3 Years RPI Average Last 3 Years FTSE 100 Average Last 3 Years 2 GDP Growth Average Last 3 Years 3 4 5 6 GDP Growth Lag 1 GDP Growth Lag 1 GDP Growth GDP Growth FTSE 100 Lag 1 RPI Lag 1 CPI FTSE 100 36 Non Macro-Economic Variables 1 0.5 0 0 2 4 6 8 10 12 14 -0.5 16 18 1 2 -1 3 -1.5 4 5 -2 6 -2.5 -3 -3.5 Incept + variable in listed order 37 Start-Up Performance logistic regression panel model panel model with year dummy panel model with selected no lagged MV (highest AIC in each category) panel model with selected one year lagged MV (highest AIC in each category) panel model with selected averaged MV (highest AIC in each category) panel model with no lagged GDP_growth rate panel model with one year lagged GDP_growth rate panel model with averaged GDP_growth rate 38 AUROC Within Sample .900 .880 .860 .840 .820 .800 .780 .760 0 1 2 3 4 5 6 models in listed order 7 8 9 10 39 Non-Start-Up Model logistic regression panel model panel model with year dummy panel model with selected no lagged MV (highest AIC in each category) panel model with selected one year lagged MV (highest AIC in each category) panel model with selected averaged MV (highest AIC in each category) panel model with no lagged GDP_growth rate panel model with one year lagged GDP_growth rate panel model with averaged GDP_growth rate 40 AUROC In Sample .900 .880 .860 .840 .820 .800 .780 0 1 2 3 4 5 6 models in listed order 7 8 9 10 41 Out-of-Sample Performance 2010 Model Non logistic regression .837 St .753 panel model .828 .757 panel model with year dummy .843 .769 panel model with selected no lagged MV (highest AIC in each category) panel model with selected one year lagged MV (highest AIC in each category) panel model with selected averaged MV (highest AIC in each category) panel model with no lagged GDP_growth rate .843 .843 .843 .758 .758 .758 panel model with one year lagged GDP_growth rate .832 .759 .758 panel model with averaged GDP_growth rate .842 .758 .833 42 Future? • Continue to explore macro-economic variables • Model based on normal • Non-parametric models • Larger range of data • Out-of-Time Sample 43 Conclusion • There is considerable stability across models - Estimates - Performance Variables • Some variables need reconsideration • GDP seems an important Macro-economic variables • BUT need further exploration 44