Jean Louis Arcand, Enrico Berkes and Ugo Panizza IMF Working paper 2011 Aliyev Namig Benlalli Yannis Paris 2012 Too much finance ?

Jean Louis Arcand, Enrico Berkes and Ugo Panizza

IMF Working paper 2011

Aliyev NamigBenlalli Yannis

Paris 2012

Too much finance ?

Content• Motivation of the paper

• Literature revue/stylized facts

• Datas and Methodology

• Results

• Extensions

• Conclusion

Motivation of the paper

•Despite the huge existing litterature, crisis has altered the mindset on the supposed positive impact of finance on growth

•Examine relationship Finance growth including non monotonicity:

•Test hypothese of "too much" finance with the existence of a threshold above which effect of financial depth becomes negative

•Use new database, GMM estimators

LiteratureFirst empirical studies :

•Goldsmith (1969), positive relationship finance and long run growth. Higher efficiency of investments

•Joan Robinson (1952) economic growth causes financial development. "Where enterprises leads, finance follows"

In the 90's studies concentrates on the causality from finance to growth:

•King and Levine (1993) financial depth predict growth, Levine and Zervos (1998) stock market liquidity positive impact on growth, Levine Loayza and Beck (2000), Rajan and Zingales (1998) industrial sector more dependent on finance.

Literature (ctd)

In the 2000's skepticism. Concern about the robustness of the finance growth nexus :

•Demetriades and law (2006), no impact with poor institutions, Rousseau and Watchel (2002), no impact with double digit inflation, Rioja and Valev (2004)

•Concerns about "too large" financial systems : Easterly Islam and Stiglitz (2000), Rajan (2005), Rousseau and Watchel (2011)

•To sum up, trade off short/long term growth. Positive impact in the long run despite higher volatility in the short run. Loayza and Ranciere (2006), Ranciere Tornell and Westermann (2008)

Data

Country and industry level data :

•69 countries over the period 1960-2010

•Data on GDP growth, credit to private sector, average years of schooling, inflation, government expenditures, institutional quality, banking supervision, regulation and monitoring indicators. From the World bank mostly

•39 industries over 1990-2000 : it includes data on value added growth in industry i for country j, external financial dependence for the US manufacturing. Rajan and Zingales (1998) indicators

Methodology

• PC : measure total credit to private sector (PC) Importance of using deposits bank and other financial institutions since 2000. Indeed in the US total credit to private sector four times larger than banking deposits.

• Yi : capture GDP growth

• Zi : set of control variables generally used in the literature => log of initial GDP per capita, initial stock of human capital, inflation, trade openess, ratio of government expenditure to GDP

Yi=a0+BPC+CPC^2+Zt+e

ResultsOmitted variable bias corrected with Quadratic term (non monotonicity)

Although it depends on the method used, negative marginal effect of financial depth when PC reaches 80-100% of GDP

Results consistent with respect :

•Different estimators: simple OLS, Panel GMM as well as Semi parametric estimators

•Different data : Country/Industry level

•Findings robust to controlling for macroeconomic volatility, banking crisis and institutional quality

ResultsComparing with other results

•Results similar to those of Stiglitz, Easterly and Islam (2000). Treshold effect on output volatility•Consistent with the "vanishing effect" found by Rousseau and Watchel (2011)•Contradict Rioja and Valev (2004), positive impact of financial depth in regions of high finance

Results : OLS regression

Results : OLS regression for more recent periods

•Use average GDP/capita growth•Positive and significant coefficient of

credit to private sector using level or log, and negative quadratic term (concave relationship). Treshold =83% for 1970-2000

•1990-2010 without quadratic term, decrease by nearly 50% of the coefficient : downward bias increasing over time•Downward and increasing bias in miss

specified model, but there is still endogeneity with OLS (reverse causality).

Panel estimation

• Use of time Variation dividing the sample into 6 non overlapping 5-year periods

• GMM method to deal with endogeneity. It includes time fixed effects, lag values of PC and all other control variables

• "Vanishing effect" of financial depth because of growing financial sector over time : between 2000-10, countries with PC/GDP>90% increased from 4% to 22% of the sample

Results : panel estimation

Results : Panel estimation

•Linear and quadratric variable statistically significant. Marginal effect of financial depth becomes negative when PC=140% for the period 1960-1995. Threshold reaches 100% for 1960-2005 and 90% for even more recent data• Results robust when excluding outliers : countries with PC>165%. Threshold reaches 69%.

Extensions

• Volatility, Crises and Heterogeneity

• Industrial level data

• Conclusions

Volatility, Crises and HeterogeneityVolatility- within country standard deviation of annual

output growth for each periodEstimated Model:GRi,t= β0PCi,t−1 + β1PCi,t−1 + (PCi,t−1b0 +PCi,t−1

b1 + δ) × HVOLi,t + Xi,t−1 Γ + αi + τ t + εi,t

HVOL- dummy variable , that =1 if volatility greater the sample avarage of 3.5, and =0 otherwise

GMM estimation Panel

(1) (2) (3) (4) LGDP(t-1) -0.356 -0.347 -0.693** -0.548*PC(t-1) 2.925* 2.999** 3.334* 3.957** PC2(t-1) -1.982** -2.104** -1.577* -2.431** HVOL -1.326*** -1.076** PC(t-1)×HVOL -1.399 PC2(t-1)×HVOL 0.868BKCR(t) -1.898*** -2.134** PC(t-1)×BKCR(t) -0.013PC2(t-1)×BKCR(t) 0.689 LEDU(t-1) 1.570** 1.726*** 2.155*** 1.871***LGC(t-1) -1.734*** -1.570*** -1.709** -1.843***LOPEN(t-1) 1.323*** 1.041*** 1.008* 0.999** LINF(t-1) -0.133 -0.032 -0.010 -0.032 Cons. -0.074 0.070 1.604 1.590N. Obs. 917 917 872 872N. Cy. 133 133 133 133 AR1 -5.12 -5.11 -4.95 -4.87 p-value 0.00 0.00 0.00 0.00 AR2 -1.34 -1.27 -1.02 -1.18 p-value 0.180 0.203 0.307 0.236OID 119.5 122.7 126.3 122.4

Period 1960-2010 1960-2010 1970-2010 1970-2010 dGR/dPC=0 0.74 0.7 1.06 0.81 dGR/dPC=0 (HV or BC) 0.65 1.13 Robust (Windmeijer) standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Volatility, Crises and Heterogeneity

Then authors substitute volatility dummy variable with a banking crisis dummy:

• BKCR -for bank crises=1 in crise periods, in tranquil periods =0(Leaven and Valencia, 2010 database)

GMM estimation Panel

(1) (2) (3) (4) LGDP(t-1) -0.356 -0.347 -0.693** -0.548*PC(t-1) 2.925* 2.999** 3.334* 3.957** PC2(t-1) -1.982** -2.104** -1.577* -2.431** HVOL -1.326*** -1.076**PC(t-1)×HVOL -1.399 PC2(t-1)×HVOL 0.868BKCR(t) -1.898*** -2.134** PC(t-1)×BKCR(t) -0.013PC2(t-1)×BKCR(t) 0.689 LEDU(t-1) 1.570** 1.726*** 2.155*** 1.871***LGC(t-1) -1.734*** -1.570*** -1.709** -1.843***LOPEN(t-1) 1.323*** 1.041*** 1.008* 0.999** LINF(t-1) -0.133 -0.032 -0.010 -0.032 Cons. -0.074 0.070 1.604 1.590N. Obs. 917 917 872 872N. Cy. 133 133 133 133 AR1 -5.12 -5.11 -4.95 -4.87 p-value 0.00 0.00 0.00 0.00 AR2 -1.34 -1.27 -1.02 -1.18 p-value 0.180 0.203 0.307 0.236OID 119.5 122.7 126.3 122.4

Period 1960-2010 1960-2010 1970-2010 1970-2010 dGR/dPC=0 0.74 0.7 1.060.81 dGR/dPC=0 (HV or BC) 0.65 1.13 Robust (Windmeijer) standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Volatility, Crises and Heterogeneity Then authors add other dummies and time invariant

variables:• LQOG-for low quality of government=1 if ICRG index

is below 0.5,(median=0.51) otherwise=0• LOSI-for bank supervision=1 in low, =0.5 in

intermediate,=0 in high bank supervision countries (time invariant variable)

• LKRI-for capital regulatory =1 with low capital stringency and otherwise=0 (time invariant variable)

• LPMI-for private monitoring index=1with low private monitoring, otherwise=0 (t.i.v.)

Institutional Quality and Bank Regulation and Supervision

(1) (2) (3) (4)

LGDP(t-1) -0.416 -0.754** -0.520* -0.607* PC(t-1) 3.443* 4.505 2.785 2.306 PC2(t-1) -2.459*** -2.797* -2.387* -1.810* LQOG(t-1) 0.386 PC(t-1)×LQOG(t-1) -1.982PC2(t-1)×LQOG(t-1) 1.249LOSI -0.746 PC(t-1)×LOSI -1.929 PC2(t-1)×LOSI 1.228LKRI -1.657 PC(t-1)×LKRI 0.188PC2(t-1)×LKRI 0.978LPMI -1.4 82 PC(t-1)×LPMI 1.300PC2(t-1)×LPMI -0.525

N. Obs 819 917 828 917 N. Cy 115 133 116 133AR1 -4.82 -5.33 -4.93 -5.34 p-value 0.00 0.00 0.00 0.00AR2 -1.47 -1.60 -1.47 -1.54 p-value 0.142 0.12 0.147 0.123OID 95.83 110.1 99.8 111.9P-value 1 1 1 1 dGR/dPC=0 0.70 0.81 0.58 0.64 dGR/dPC=0 INT 0.60 0.82 1.05 0.77 Robust (Windmeijer) standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1

Indusrty level data

Estimated Model is following:VAGRi,j =α SHVAi,j + EFj × (βPCi + γPC2i ) + λj + µi

VAGRi,j- real value added growth of industry j in country i over the 1990-2000 period

SHVAi,j-initial share of value added of industry j over total industrial value added in country i

Efj-index of external financial dependance for industry j in the 1990s (R&Z index)

λj and µi- a set of industry and country fixed effects

Rajan and Zingales Estimations (1) (2) (3) (4) (5) (6) SHVAt−1 -2.069** -2.059** -2.063** -2.061** -0.645 -2.217** EF×PC 0.0180* 0.0742** 0.0696** 0.0654* 0.0508** EF PC2 -0.0300** -0.0284** -0.0265* -0.0227* EF×Y 0.000945 0.0309EF×Y2 -0.00181OEF×PC 0.169*** OEF×PC2 -0.0694***

Constant 0.0648*** 0.0681*** 0.0691*** 0.0869*** 0.0508*** 0.0510** PC thresh. 1.237 1.225 1.234 1.119 1.218 N. Obs. 1252 1252 1252 1252 1252 1252 R-squared 0.3360.338 0.338 0.338 0.433 0.343 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Conclusions

• There is a non-monotone realtionship between financial depth and economics growth

• There is a certain threshold-around 80-100% of GDP-above which finance starts having negative effect on economic growth

Thank you!!!