Non-Performing Assets and Lending Growthbombaychamber.com/admin/uploaded/Reference Material... · I...
Transcript of Non-Performing Assets and Lending Growthbombaychamber.com/admin/uploaded/Reference Material... · I...
Non-Performing Assets and Lending Growth
Amartya Lahiri* Urvi Neelakantan†
*CAFRAL and University of British Columbia
†CAFRAL and FRB Richmond
March 2019
1 / 33
Questions
I Has bank lending followed standard profitability criteria?
I Have NPAs been driven by borrower productivity?
I How are loans, productivity and NPAs related?
2 / 33
This Presentation
I Examine questions at the sectoral level
I Use variation in loans, NPAs and total factor productivity(TFP)
I cross-sectional variation across 19 sectors
I time-series variation between 1999-2018
3 / 33
Data Sources
I Bank Statistical Returns (BSR) data
I amount outstanding, by sector, between 1999-2018
I outstanding loans broken into four groups
I groups 2-4 classified as NPA
I Indian KLEMS data
I sectoral output, input and TFP data
I covers 1981-2015
I PROWESS database for sectoral q
4 / 33
Pattern 1: Time trend of NPA
Figure: What fraction of each sector’s loans are NPAs?
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
50%
1999 2004 2009 2014
Basic Metals
Vehicles
Engineering
Leather
Construction
Agriculture
1999 Average: 12.9% 2018 Average: 7.8%
Mining
5 / 33
Pattern 2: Sectoral Share of NPAs in 2018I Four sectors currently account for half of all NPAs: Metals,
Construction, Agriculture and Electricity/Gas/Water
Basic Metals & Metal Products
18%
Construction16%
Agriculture -Direct Finance
8%Electricity, Gas &
Water8%
Professional and Other Services
6%
Engineering5%
Textiles5%
Wholesale Trade5%
Food Manufacturing &
Processing4%
Retail Trade3%
Other Industries3%
Vehicles,Vehicle Parts & Transport
Equipments3%
6 / 33
Pattern 3: Loan share v NPA share in 2018
Figure: Sector Share of Total Lending vs. Sector Share of Total NPAs
Transport Operators
FinanceMining
Textiles
Metals
Engineering
Vehicles
Electricity, Gas & Water
Construction
Personal Loans -Housing
Personal Loans -Others
0%
2%
4%
6%
8%
10%
12%
14%
16%
18%
20%
0% 2% 4% 6% 8% 10% 12% 14%
Sect
or S
hare
of T
otal
NPA
s (20
18)
Sector Share of Total Lending (2018)
NPA share 30% greater
NPA share=Loan Share
NPA share 30% lower
7 / 33
Pattern 4: Sectoral NPAs – Then and NowWhat fraction of each sector’s loans are NPAs (1999 vs. 2018)?
Sector NPA Share Sector NPA Share
1999 2018 1999 2018
Basic Metals & Metal Products 17% 45% Professional and Other Services 13% 10%
Vehicles, Vehicle Parts & Transport Equipment 6% 32% Leather & Leather Products 25% 9%
Engineering 15% 28% All Others 13% 9%
Mining & Quarrying 9% 25% Rubber & Plastic Products 12% 8%
Construction 21% 24% Agriculture - Direct Finance 18% 8%
Paper, Paper Products & Printing 16% 24% Petroleum, Coal Products & Nuclear Fuels 5% 8%
Textiles 17% 21% Retail Trade 10% 8%
Food Manufacturing & Processing 16% 20% Transport Operators 13% 6%
Electricity, Gas & Water 3% 17% Agriculture - Indirect Finance 7% 6%
Other Industries 10% 17% Personal Loans - Consumer Durables 5% 4%
Manufacture of Cement & Cement Products 13% 14% Personal Loans - Others 6% 3%
Beverage & Tobacco 13% 13% Personal Loans - Housing 4% 2%
Chemicals & Chemical Products 17% 12% Finance 5% 1%
Wholesale Trade 6% 11%
8 / 33
Profit-Based Lending
I Sectors with high TFP should have higher profits
I Profit maximizing lenders should lend more to sectors withhigher profits
⇒ Sectoral loans and TFP should be positively correlated
I High profit sectors should have lower NPAs
⇒ Sectoral TFP and NPAs should be negatively correlated
I Does this show up in the data?
9 / 33
Aggregate DataCorrelations
NPA Share Loan Growth TFP Growth
NPA Share 1
Loan Growth -0.1957 1
TFP Growth -0.5349** 0.5102* 1
10 / 33
Lending and Productivity: Sectoral DataExpect Positive Correlations
Figure: Lead-Lag Correlation Between TFP Growth and Lending GrowthRates (1999-2015)
-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5 C
onst
ruct
ion
Min
ing
& Q
uarr
ying
Bas
ic M
etal
s &
Met
al P
rodu
cts
Tra
de
Pet
role
um, C
oal P
rodu
cts &
…
Veh
icle
s,Ve
hicl
e Pa
rts
&…
Cem
ent &
Cem
ent P
rodu
cts
Foo
d Be
vera
ge a
nd T
obac
co
Eng
inee
ring
Rub
ber &
Pla
stic
Pro
duct
s
Ele
ctric
ity, G
as &
Wat
er
Woo
d, G
ems,
Oth
er In
dust
ries
Tex
tiles
& L
eath
er
Che
mic
als
& C
hem
ical
Pro
duct
s
Agr
icul
ture
Tra
nspo
rt O
pera
tors
Pap
er, P
aper
Pro
duct
s &
…
Pro
fess
iona
l & O
ther
Ser
vice
s
Fin
ance
-3 -2 -1 0 1 2 3
Note: TFP and lending indexed to 100 in 1999. Certain KLEMS sectors were combined to align with BSR sectorcodes. TFP Growth is anchor variable, Lending Growth Rate’s lag and lead are taken.
11 / 33
Productivity and NPA: Sectoral DataExpect Negative Correlations
Figure: Lead-Lag Correlation Between TFP Growth and Fraction ofSector Loans that are NPAs (1999-2015)
-1.2-1
-0.8-0.6-0.4-0.2
00.20.40.60.8
1
Eng
inee
ring
Che
mic
als
& C
hem
ical
…
Fin
ance
Pap
er, P
aper
Pro
duct
s &
…
Tra
nspo
rt O
pera
tors
Rub
ber &
Pla
stic
Pro
duct
s
Agr
icul
ture
Ele
ctric
ity, G
as &
Wat
er
Pro
fess
iona
l & O
ther
…
Foo
d Be
vera
ge a
nd…
Tex
tiles
& L
eath
er
Veh
icle
s,Ve
hicl
e Pa
rts
&…
Cem
ent &
Cem
ent…
Tra
de
Pet
role
um, C
oal P
rodu
cts…
Woo
d, G
ems,
Oth
er…
Min
ing
& Q
uarr
ying
Con
stru
ctio
n
Bas
ic M
etal
s &
Met
al…
-3 -2 -1 0 1 2 3
Note: TFP and lending indexed to 100 in 1999. Certain KLEMS sectors were combined to align with BSR sectorcodes. TFP Growth is Anchor variable, NPA share in total sector’s lag and lead are taken
12 / 33
Summary of Sectoral Data Patterns
I Basic Metals, Construction and Mining are commonaberrations
I Collectively account for 36 percent of NPAs
I Account for only 13 percent of loans outstanding
I Are these sectors statistical outliers?
13 / 33
Econometric Testing
I Formally examine relationship between loans, productivity andNPAs
I Which sectors are econometric outliers?
14 / 33
Model of Lending
I Baseline specification
∆Lit = F
(NPAit−1
NPAt−1, qit−1,
Dit−1
Ait−1
)+ εit
I q = market valuebook value is Tobin’s q
I Dit−1
Ait−1is debt-to-asset ratio
I Problem: NPA respond to q and debt-to-asset ratio
15 / 33
Two-Step Process
I Step 1: Estimate
NPAit
NPAt= G
(Lit−1
Lt−1, qit−1,
Dit−1
Ait−1,CPIt−1
)+ ηit
I Step 2: Estimate
∆Lit = F
(η̂it−1, qit−1,
Dit−1
Ait−1, t
)+ εit
I CPI is commodity price index
I η̂it measures NPAs not due to other variables
16 / 33
Null Hypothesis
I Expect coefficient on loan share to be one
I if NPAs are random then expected NPAs are zero
I sectors with higher loan shares will have higher NPA shares
I Coefficient below one indicates banks doing better thanrandomly allocating loans
I Coefficient above one indicates coin toss would do better
17 / 33
First-step Result: Sector Share of NPAs(1) (2) (3) (4)
Sector Share 0.863∗∗∗ 0.835∗∗∗ 0.657∗∗∗ 0.612∗∗∗
of Loans Lagged (18.37) (18.65) (10.36) (9.09)
Debt-to-Asset 0.0912∗∗∗ 0.0331∗∗
Ratio Lagged (7.68) (1.98)
Tobin’s Q 0.00542∗∗∗ -0.000457Lagged (3.69) (-0.31)
Comm. Price -0.0000185 0.00000538Index Lagged (-0.41) (0.15)
Sector No No Yes YesFixed Effects (.) (.)
Constant 0.00794∗∗∗ -0.0264∗∗∗ 0.00588 -0.00346(3.57) (-4.45) (1.25) (-0.47)
Observations 418 418 418 418R2 0.448 0.520 0.737 0.739
t statistics in parentheses∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
18 / 33
Actual and Predicted Share of NPAs by Sector
0.1
.2.3
0.1
.2.3
0.1
.2.3
0.1
.2.3
0.1
.2.3
2000 2005 2010 2015 2020 2000 2005 2010 2015 2020 2000 2005 2010 2015 2020
2000 2005 2010 2015 2020 2000 2005 2010 2015 2020
Transport Operators Prof & Other Services Finance Agriculture Mining & Quarrying
Food Mfg & Proc Beverage & Tobacco Textiles Paper & Printing Leather
Rubber & Plastic Chemicals Petrol, Coal, Nuclear Cement Metals
Engineering Vehicle & Trans Equip Wood, Gems, Other Electricity, Gas, Water Construction
Wholesale Trade Retail Trade
Actual Predicted
Sect
or S
hare
of E
cono
my
NPA
s
year
Graphs by bsr_ind_code
19 / 33
Actual and Predicted Share of NPAs: Mining
0.0
05.0
1.0
15.0
2Se
ctor
Sha
re o
f Eco
nom
y N
PAs
2000 2005 2010 2015 2020year
Actual Predicted
20 / 33
Actual and Predicted Share of NPAs: Metals
.05
.1.1
5.2
.25
Sect
or S
hare
of E
cono
my
NPA
s
2000 2005 2010 2015 2020year
Actual Predicted
21 / 33
Actual and Predicted Share of NPAs: Vehicles
0.0
1.0
2.0
3.0
4Se
ctor
Sha
re o
f Eco
nom
y N
PAs
2000 2005 2010 2015 2020year
Actual Predicted
22 / 33
Actual and Predicted Share of NPAs: Electricity, Gas,Water
0.0
2.0
4.0
6.0
8Se
ctor
Sha
re o
f Eco
nom
y N
PAs
2000 2005 2010 2015 2020year
Actual Predicted
23 / 33
Actual and Predicted Share of NPAs: Construction
0.0
5.1
.15
Sect
or S
hare
of E
cono
my
NPA
s
2000 2005 2010 2015 2020year
Actual Predicted
24 / 33
Second-step Result
Lending Growth
Residual -0.560∗
Lagged (-1.89)
Tobin’s Q 0.00742Lagged (0.84)
Debt-to-Asset -0.250∗∗∗
Ratio Lagged (-2.63)
Year YesFixed Effects
Sector YesFixed Effects
Observations 396R2 0.371
t statistics in parentheses∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
25 / 33
Actual and Predicted Lending Growth
-.50
.5-.5
0.5
-.50
.5-.5
0.5
-.50
.5
2000 2005 2010 2015 2020 2000 2005 2010 2015 2020 2000 2005 2010 2015 2020
2000 2005 2010 2015 2020 2000 2005 2010 2015 2020
Transport Operators Prof & Other Services Finance Agriculture Mining & Quarrying
Food Mfg & Proc Beverage & Tobacco Textiles Paper & Printing Leather
Rubber & Plastic Chemicals Petrol, Coal, Nuclear Cement Metals
Engineering Vehicle & Trans Equip Wood, Gems, Other Electricity, Gas, Water Construction
Wholesale Trade Retail Trade
Actual Predicted
Lend
ing
Gro
wth
year
Graphs by bsr_ind_code
26 / 33
Conclusion
I NPAs have risen secularly since 2011, and spiked over the pastthree years
I A few sectors have been outliers, construction and metalsbeing two such
I NPAs of outliers have exceeded the amount that fundamentalscan explain
I Banks, on average, did well: sectoral NPA shares < sectoralloan shares
I Problem appears to be specific misallocation, not generalinefficiency
27 / 33
Sector Share of NPAs and Sector Share of Risky Firms
Sector Share of NPAs Debt-to-Asset Ratio Sector Share of Risky FirmsSector Share of NPAs 1.0000Debt-to-Asset Ratio 0.3562*** 1.0000Sector Share of Risky Firms 0.5035*** 0.1757*** 1.0000
Risky firms refers to firms in the top quartile of fitted probability of default.
28 / 33
Probability of Default Series: MethodologyA. Sample ConstructionI Data regarding listed firms characteristics and defaults is
collected for the period 2002 – 2018 as follows:I A1. Firm characteristics data : Accounting and stock
market data gives a daily panel of all listed firmsI A2. Firm level default data : Sample of firms that have
defaulted on their debt payment obligations in each quarter
I A ‘quarterly default dummy’ is created, which is set to 0 if fora given quarter a firm appears in the characteristic datasetbut not the default dataset.If a firm appears in both thedatasets, the variable is set to 1.
B. Variable DescriptionI B1. Dependent Variable: YDD (Yearly Default Dummy)
I If for firm i in year t, the sum of ‘quarterly default dummy’from Q1 to Q4 is greater than 0, then YDDit is 1, else 0.
I After the first occurrence of YDD being 1 for a company, allsubsequent observations for the company are dropped.
29 / 33
Probability of Default Series: Methodology
I B2. Independent VariablesThe independent variables are calculated using the datasetdescribed in A1. For obtaining an annual measure of each ofthese variables, the value is taken as of the 1st of Jan forcompany i in year t in the yearly panel.I Distance to Default (DTD) : Using the KMV Merton
distance-to-default methodologyI Interest Expense to PBDITA (Int2PB) : Interest Expense
PBDITAI Book-to- Market Ratio (Bk2Mkt) :
Total AssetsMarket Capitalization +Total Liabilities
I Liquidity (Lqdty) : Cash and Cash EquivalentsTotal Assets
I Profitability (Pft) : Net IncomeTotal Assets
30 / 33
Probability of Default Series: Methodology
C. Logistic Regression
I For obtaining the coefficients required for calculating thefitted PDs for year t, a rolling regression of the following formis estimated on the yearly panel over the last four years i.e.over window r ranging from t-4 to t-1:
YDDr = α+β1DTDr +β2Int2PBr +β3Bk2Mktr +β4Lqdtyr+
β5Pftr + εr ....(1)
I Thus, for the period 2002 to 2018, we get 13 coefficients (onefor each year, starting in 2006 and ending in 2018).
31 / 33
Probability of Default Series: MethodologyD. Daily Fitted PDsBase data is the daily panel of listed firms as created in A1. Thedaily fitted PD are calculated as follows:I For each firm i for each day j in a given year t, the Sum is
calculated using the coefficients obtained in equation (1) asfollows:
Sumitj = α+β1DTDitj+β2Int2PBitj+β3Bk2Mktitj+β4Lqdtyitj
+β5Pftitj ....(2)
I For each day j, the fitted PD (FittedPD) is calculated byusing Sumitj in a logistic function as follows:
FittedPDj =eSumitj
1 + eSumitj
E. Annual Fitted PDsFor each firm i in a given year t, we obtain the annual fitted PDvalue by taking an average across the daily values in that year.
32 / 33
Measure Of Riskiness : MethodologyA.Classification Of Firms Based on RiskI Based on the distribution of fitted PD variable, a threshold
value is defined to classify each of the firms as high risk or lowrisk. For a firm i in a year t, if the fitted PD value is greaterthan the threshold, then the variable ‘high risk dummy’(HRDit) takes the value 1, otherwise 0.
I Two different ‘high risk dummy’ variables are generated usingthe threshold as 75th and 80th percentile value of fitted PDrespectively.
B.Calculating Industry RiskTo generate the risk measure at an industry level, we define a newvariable which is calculated as follows:
I For an industry m in year t,
RMmt =No of firms in industry m with HRDit = 1
Total no. of firms with HRDit = 1
33 / 33