Unbelievable Balance Sheets: Predicting banks with Hidden

9th Biennial Conference, Czech Economic Society
Unbelievable Balance Sheets:
Predicting banks with Hidden Negative Capital
(preliminary results)
Mikhail Mamonov
Center for Macroeconomic Analysis and Short-term Forecasting (CMASF),
Institute for Economic Forecasting RAS,
National Research University “Higher School of Economics”
Prague
November 26, 2016
Data on Russian banks: A new Head of the Bank of Russia …
2
1050
New Head:
Elvira Nabiullina
1000
950
900
850
Previous Head:
Sergey Ignatiev
18
16
14
12
10
8
750
6
700
4
650
2
600
0
2008M1
2008M5
2008M9
2009M1
2009M5
2009M9
2010M1
2010M5
2010M9
2011M1
2011M5
2011M9
2012M1
2012M5
2012M9
2013M1
2013M5
2013M9
2014M1
2014M5
2014M9
2015M1
2015M5
2015M9
2016M1
2016M5
800
Number of bank failures (right axxis)
Number of banks disclosing balance sheets through the Bank of Russia web-site
Approx. every 3rd bank was banned from the banking system
in just 3 years (!)
Data on Russian banks: fraud attacks!
30
20
0
10
Frequency
40
50
3
0
10
20
30
NNW-to-EQ
NNW-to-EQ = The ratio of absolute value of Negative Net Worth (revealed after the license
withdrawal) to positive value of capital (reported before the license withdrawal), times
Preliminary statistics: “champions” in negative net worth
(top-20 out of 106 cases over the 2013M1-2016M1)
4
№
Bank name
1
Mosoblbank
2
National Bank "TRUST"
3
Vneshprombank
4
Rossiysky Kredit
5
Interkommerz Bank
6
Investbank
7
Probiznesbank
8 SB Bank (Shipbuilding bank)
9
Nota-Bank
10 Mosstroyeconombank (М Bank)
11
Bank Transportny
12
Master-bank
13 Pervy Respublikansky Bank
14
Tusar
15
Narodny Kredit
16
Bank Fininvest
17
Zapadny
18
Ogni Moskvy
19
Klientsky
20
Pushkino
Total (in RUB bln):
Date of withdrawal
/ sanation*
May-14*
December-14*
January-16
July-15
February-16
December-13
August-15
February-15
November-15
July-15
May-15
November-13
May-14
September-15
October-14
July-14
April-14
May-14
July-15
September-13
Total (% of last reported assets):
Net worth The last positive
Assets,
City
(ex post), value of capital (ex
bln. rub.
bln. rub.
ante),bln. rub.
Moscow
–172.0
17.8
78.3
Moscow
–129.0
16.2
302.6
Moscow
–210.1
16.1
293.7
Moscow
–111.0
17.4
186.7
Moscow
–60.0
6.9
110.0
Moscow
–44.5
6.1
78.0
Moscow
–40.8
9.7
154.5
Moscow
–39.1
7.6
82.9
Moscow
–35.5
10.9
78.0
Moscow
–28.3
3.4
49.0
Moscow
–21.4
3.0
52.7
Moscow
–17.2
9.1
85.8
Moscow
–16.2
3.7
40.5
Moscow
–15.0
1.5
19.6
Moscow
–12.7
3.7
41.0
St. Petersb.
–12.7
1.7
20.1
Moscow
–12.2
2.4
31.2
Moscow
–12.2
1.3
21.5
Moscow
–11.4
1.6
17.8
Pushkino
–10.9
2.8
31.0
–1031.4
142.9
–58%
8%
1775.0
The problem of ex post NNW of banks: only in Russia?

5

According to the Bank of Russia press releases over the 2013М6 – 2016М10:

284 bank licenses were withdrawn and at least 3 large banks were sanated,

In 183 cases the NNW was revealed,

After the withdrawals Total NNW= -1683 bln RUB (-2.1% of GDP or -47% of
failed banks’ assets),

Before the withdrawals Total capital = +327 bln RUB (+0.4% of GDP or +9% of
failed banks’ assets),

Reasons of NNW accumulation: unsuccessful aggressive lending and fraud.
This is not a story of weak institutions in emerging markets as compared to
developed economies. For example, in US over the 2007 – 2013:

403 banks were closed by the FDIC,

Total NNW = -24% of failed banks’ assets (Balla et al., 2015; Cole, White, 2015),

Before the withdrawals Total capital = +1.5% of failed banks’ assets,

Main reasons are the same (!)

Less dramatic, but qualitatively the same story as in Russia
The same problem occurred during the crisis of 1980s (James, 1991 JF)
Motivation: Growing negative net worth and fraud in banking 1/2
6






Bank failures destruct relationship lending and decrease total supply of credit to
the economy (Bernanke, Blinder, 1992, AER; Ashcraft, 2005, AER)
Bank failures with NNW accumulation creates additional reputational risks for
banking system – lack of depositors’ confidence and reduced incentives of
potential investors to buy a stake in bank’s capital
Empirical literature on NNW has been developed for US since 1980-s; however,
for emerging markets the research is yet to be done
Moreover, the literature exploits simple indicators (asset structure, liability
structure, size & NPLs) to identify the cases of NNW. But, is it enough?
The NNW is related to balance sheet falsifications. The falsification schemes are
permanently and very rapidly developed becoming more complex…
Thus, we need to exploit the existing practical evidence on schemes of hiding
NNW by bank – to propose suitable indicators describing this evidence (complex
indicators) and include them in forecasting models…
Motivation: Two forecasting exercises
2/2
7


First exercise: How to forecast the size of NNW if a bank has already failed?

Work with the sample of failed banks with non-zero NNWs only (“1”)

Try to forecast the size of NNW using Simple and Complex indicators with some
time lag (1, 3, 6, 9, 12 months)
Second exercise: How to forecast the size of NNW if a bank has not failed yet?

Work with both sample of “1” and sample of still operating banks “0”

Sample selection bias: observe NNWs in “1” only, but some of banks in “1”
might already have non-zero NNWs and hide them from the regulator

Try to forecast the number of banks in “1” with non-zero hidden NNWs and the
expected size of their NNWs (Heckman selection model)
Theoretical roots…
8

Literature on net worth of firms

Bernanke, Blinder (1989 AER): unanticipated drop in asset prices
(debt-deflation shock) – devaluation of collateral – rise of interest rates
on loans (“costly state verification” problem) – decrease in net worth of
firms

Kioytaki, Moore (1997 JPE): a small and temporary productivity shock
or income distribution shock – large and persistent drop in asset prices
and aggregated output (spiral effect). Firms’ net worth is permanently
decreased

Chen (2001 JME): introduced banks in a model a-la Kioytaki and Moore
(1997). The same negative shocks are applied – devaluation of collateral
– decrease in loans to the economy – banks’ net worth decreases (less
assets given the same level of liabilities)

Firms and banks do not hide (falsify) their accounts…
Empirical literature review: in general
9


Studies on bank failures

a bunch of papers for US (DeYoung, Torna, 2013, JFI; Cole, White, 2012,
JFSR; Clearly, Hebb, 2016, JBF; among many others)

emerging markets (Arena, 2008, JBF; Mannasoo, Mayes, 2010, JBF;
Fungacova, Weill, 2013, ET)

some for EU (Poghosyan, Cihak, 2011, JFSR; Betz et al., 2014, JBF)
Meanwhile, the literature on NNW (negative net worth)

is much less developed and only for US (James, 1991, JF; Wheelock,
Wilson, 2000, RES; Schaeck, 2008, JFSR; Bennet, Unal, 2014, JFS; Granja
et al., 2014; Balla et al., 2015; Cole, White, 2015; Kang et al., 2015, RFS)

Use simple determinants (asset structure, liability structure, capital and
risks)

no studies for emerging markets revealed.
Empirical literature review: US experience
10
«Holes» in the capital (NNW): Simple indicators outline
(“+” or “–” denotes positive or negative correlation with the size of NNW)
James
(1991)
Osterberg,
Thomson
(1995)
Schaeck
(2008)
Bank size
Bennet,
Unal
(2014)
+
Capital-to-assets
+
–
–
NPLs ratio
+
+
+
+
Uncollected income
+
+
+
+
Core liabilities
–
–
Brokered deposits
+
+
+
Deposits maturity
+/–
Types of assets (real estate, etc.)
+
Cole,
White
(2015)
+
–
+
+
+
–
Assets’ growth rate
Off-balance sheet operations
Balla et al.
(2015)
+
–
+
+
+
+
Number of branches
–
Loans to insiders
+
+
+
Data on Russian banks
11
1. Regular: The Central Bank of Russia web-site (www.cbr.ru)
 monthly balance sheets of banks (Form 101);
 quarterly profit and loss accounts (Form 102);
 monthly information on regulatory normatives (Form 135), includes,
among others, data on banks’ investments in close-end mutual funds;
2. Irregular (had-collected): press releases of the CBR (Vestnik Banka
Rossii) for data on negative net worth of failed banks estimated one
quarter after failures
3. Time period: 2013M6 – 2016M9 (since the Nabiullina’s appointment as
the Head of the Bank of Russia)
4. Number of banks: 928 at the beginning to 633 at the end of the sample
12
Estimation results: two forecasting exercises
First exercise: How to forecast the size of NNW
if a bank has already failed?
(sample of failed banks with non-zero NNW – “1”,
Estimation method: robust OLS)
Empirical design: (1) No sample selection bias accounted for /
no out-of-sample forecasting allowed
13
1. Weighted OLS: subsample of “1” (failed banks with NNW>0)
First step: Simple indicators of hidden negative capital (General 𝐵𝑆𝐹)
Second step: Simple + Complex (new) indicators of hidden negative capital
𝑵𝑵𝑾𝒊𝒕 = 𝒇 𝑺𝒊𝒎𝒑𝒍𝒆 𝑩𝑺𝑭𝒊𝒕−𝒌 , 𝑪𝒐𝒑𝒍𝒆𝒙 𝑩𝑺𝑭𝒊𝒕−𝒌 + 𝜺𝒊𝒕
where 𝑡 is failure-specific for the subsample of failed banks (“1”)
𝑁𝑁𝑊𝑖𝑡 is negative net worth, or hidden negative capital, that was revealed
by the CBR one quarter after each failure (% of disclosed capital before
each failure).
Both 𝐵𝑆𝐹𝑖𝑡−𝑘 are bank-specific (as in Cole, White, 2015, as opposed to Balla
et al., 2015) controls with 𝑡 − 𝑘 monthly lag (𝑘 = 0,1,2 … 6)
𝜀𝑖𝑡 is (normal) regression error
NNW-to-EQ = f(Simple indicators) + e
Determinants, lag = 3 months before the closure
Группа №1: структура активов
1 Corporate loans (% of TA)
14
I
II
III
IV
0.026***
(0.014)
0.032***
(0.011)
0.027*
(0.015)
0.034***
(0.011)
0.087
(0.060)
0.135**
(0.061)
2 Private securities (% of TA)
Группа №2: структура пассивов
3 Capital (% of TA)
–0.077**
(0.033)
4 Retail deposits (% of TA)
0.032***
(0.012)
0.045***
(0.010)
0.034***
(0.012)
0.045***
(0.010)
14.654***
(3.464)
15.671***
(3.255)
14.393***
(3.435)
14.857***
(3.159)
–0.070**
(0.034)
Группа №3: размер и риски
5 Size
6 Retail NPL (% of TA)
–0.420*
(0.248)
7 Assets growth (12 m., lag=12), %
Obs
–0.399
(0.251)
0.018***
(0.006)
0.018***
(0.006)
89
89
89
89
0.378
0.368
0.384
0.382
Min: Actual / Fitted
0.1/–1.8
0.1/–0.5
0.1/–1.8
0.1/–0.7
Med: Actual / Fitted
3.4/4.2
3.4/4.0
3.4/4.2
3.4/4.0
Max: Actual / Fitted
11.1/8.0
11.1/8.8
11.1/7.9
11.1/8.5
R2
Simple vs Complex indicators of NNWs: New hypotheses introduced
15
Question: Have the literature fully covered possible indicators of NNWs?
𝑆𝑖𝑚𝑝𝑙𝑒 𝐵𝑆𝐹 are bank-specific indicators of NNWs from previous research on
US: asset structure, liability structure, others (size, NPLs)
𝐶𝑜𝑚𝑝𝑙𝑒𝑥 𝐵𝑆𝐹 are bank-specific indicators that might also be relevant:
Hypotheses:
H1: Balance sheet falsification (very high reserve assets ratio, very large
credit to non-resident banks with very low turnovers, very small NPL
and very high LLP; very high close-end mutual funds-to-assets)
H2: Very high turnovers on different assets (loans to companies,
households, correspondent accounts in CBR or abroad, securities, …)
H3: Very low business margins (high reliance on retail deposits and
specialization on corporate loans; low equity growth rate and
specialization on corporate loans)
Regression results: Comparative estimates of economic effects, %*
Turnobers of corporate loans
Higher reliance on retail deposits and Higher specialization on…
Retail deposits-to-assets ratio
Bank size
16
Corporate loans-to-assets ratio
Higher specialization on corporate loans
Higher reliance on retail deposits
Turnovers of reserve assets
Annual assets growth rate
Turnovers of credit to nonresident banks
High credit to nonresident banks and low turnovers of these operations
Turnovers of private securities
High securities-to-assets ratio
Private securities-to-assets ratio
Very high close-end mutual funds-to-assets ratio
Turnovers of government bonds
Turnovers of retail NPLs
Turnovers of corporate NPLs
Turnovers of retail loans
Turnovers of cash (monthly)
Very high reserve assets ratio
Low NPLs raion and High LLPs ratio
Retail NPLs ratio
Very High corporate loans ratio and low capital growth
High corporate loans ratio and low capital growth
Capital-to-assets
-0,3
Simple indicators
Falsifications (Н1)
High Turnovers (Н2)
Low Business Margins (Н3)
-0,2
-0,1
0,0
0,1
0,2
0,3
0,4
* Notes:
1. Each NNW indicator was increased by its 1 standard deviation;
2. These increases were then multiplied by the OLS-estimates of respective coefficients to obtain
the economic effect on NNW;
3. These economic effects are then were taken as a % of NNW standard deviation
0,5
17
Estimation results: two forecasting exercises
Second exercise: How to forecast the size of
NNW if a bank has not failed yet?
(sample of failed banks with non-zero NNW – “1”
and sample of still operating banks “0” ,
Estimation method: Two-step Heckman selection)
Empirical design: (2) Sample selection bias considered /
out-of-sample forecasting possible
18
2. Heckman model: subsamples “1” and “0” combined
hidden negative capital conditional on bank failure
Regression eq.: 𝑵𝑵𝑾𝒊𝒕 = 𝒇 𝑺𝒊𝒎𝒑𝒍𝒆 𝑩𝑺𝑭𝒊𝒕−𝒌 , 𝑪𝒐𝒎𝒑𝒍𝒆𝒙 𝑩𝑺𝑭𝒊𝒕−𝒌 + 𝜺𝟏,𝒊𝒕
𝑨𝑼𝑮𝑴
∗
Selection eq.: 𝑫𝒊𝒕
= 𝒇 𝑺𝒊𝒎𝒑𝒍𝒆 𝑩𝑺𝑭𝑨𝑼𝑮𝑴
+ 𝜺𝟐,𝒊𝒕 > 𝟎
𝒊𝒕−𝒌 , 𝑪𝒐𝒎𝒑𝒍𝒆𝒙 𝑩𝑺𝑭𝒊𝒕−𝒌
𝐴𝑈𝐺𝑀
𝐴𝑈𝐺𝑀
𝑆𝑖𝑚𝑝𝑙𝑒 𝐵𝑆𝐹𝑖𝑡−𝑘
and 𝐶𝑜𝑚𝑝𝑙𝑒𝑥 𝐵𝑆𝐹𝑖𝑡−𝑘
are augmented BSF (incl. those
that do not affect the Regression Eq., but affect the Selection Eq.)
𝐶𝑜𝑟𝑟 𝜀1,𝑖𝑡 , 𝜀1,𝑖𝑡 = 𝜎12 is correlation between regressions errors:


if 𝜎12 ≠ 0 and statistically significant, than the sample selection problem
does matter so that either OLS, Quantile or Tobit will be biased;
if 𝜎12 = 0, than there is no selection problem and the choice between
OLS, Quantile, Tobit and Heckman is the matter of goodness of fit
Heckman estimation results: In-sample fit (as of February 1, 2016)
(subsamples of “1”s and “0”s are jointly considered)
Group of variables (lag = 3M):
Name
Basic model
Selec Eq.
Extended model
Reg Eq.
Selec Eq.
Reg Eq.
Simple
19 determinants:
Group 1: Asset structure
Group 2: Liability structure
Group 3: Size & Risks
1. Corporate loans (% of TA)
0.022***
2. Private securities (% of TA)
0.152***
0.155*
0.140***
1. Capital (% of TA)
–0.006
0.003
–0.009
–0.071**
2. Retail deposits (% of TA)
0.011***
0.066***
0.010***
0.001*
1. Size
–0.370
8.318**
–0.602
7.582**
2. Retail NPL (% of TA)
0.053*
–0.121
–0.030
–0.218**
1. ROE adj < SYS
–0.111*
–0.509**
Loans < 25% of TA
–0.035**
–0.102
1. T. on corresp. acc. with the Bank of Russia /100
0.000
0.120
2. T. on private securities
0.001
0.003**
3. T. on corporate loans
–0.088
3.501**
4. T. on corporate NPL
0.026*
0.023
Complex determinants:
H1: Falsifications
H2: High Turnovers
H3: Low Business Margins
1. Ret Dep > SYS × Corporate Loans > SYS /100
0.138***
2. Corporate Loans >75% of TA × ROEadj < SYS
Intercept
–0.006***
–2.319***
–1.313***
Cens / Uncens Obs
686/ 106
685/ 105
Log L
3.3 / 3.8
3.3 / 3.4
0.14
0.56***
Corr(e1,e2)
In-sample forecasts of NNWs
(Heckman selection model, 2013M6 – 2015M12)
Basic Model
Extended Model
Subsample
Subsample
20
“1”
“0”
“1”
“0”
122
610
120
579
1.37%
36.94%
1.60%
19.04%
111
461
115
508
1.31%
6.94%
1.54%
9.89%
94
323
100
365
1.12%
3.46%
1.48%
6.95%
123
697
123
697
1.37%
0%
1.37%
0%
Probability Threshold = 1%
Number of banks with NNWs > 0
Total NNWs, % of GDP 2015
Probability Threshold = 5%
Number of banks with NNWs > 0
Total NNWs, % of GDP 2015
Probability Threshold = 10%
Number of banks with NNWs > 0
Total NNWs, % of GDP 2015
Actual data (to compare with in-sample)
Number of banks with NNWs > 0
Total NNWs, % of GDP 2015
I. Actual data
II. Forecasts by the model:
Basic
Extended
Actual data:
(A) July 1, 2015: Already realized cases
21
Number of banks with HNC>0
Total HNC, % of GDP 2015
76
0.83%
(B) January 1, 2016: Already realized cases
Number of banks with HNC>0
Total HNC, % of GDP 2015
122
1.78%
Out-of-sample forecasts
of NNW
(Heckman selection model,
2013M6 – 2015M6)
(B) – (A): to be out-of-sample forecasted
Number of banks with HNC>0
46
Total HNC, % of GDP 2015
0.95%
Out-of-sample forecast as of July 1, 2015: How many banks failed during the next 6M we have predicted?
Probability Threshold = 1%
Number of banks with NNWs > 0
Total NNWs, % of GDP 2015
41
38
0.89%
0.93%
32
28
0.85%
0.81%
22
16
0.27%
0.56%
Probability Threshold = 5%
Number of banks with NNWs > 0
Total NNWs, % of GDP 2015
Probability Threshold = 10%
Number of banks with NNWs > 0
Total NNWs, % of GDP 2015
Robustness checks / Directions of future work
22
I. Checking the robustness of the findings by

re-estimating OLS and Heckman models
 with additional set of bank-level controls;
 for each month within the Nabiullina’s era in the Bank of Russia
(within feasible interval – January 1, 2014 to October 1, 2016)
II. Improving the out-of-sample forecasts performance of Heckman
models
III. Credit slowdown or even crunch ? What are the real effects of negative
net worth ?
IV. How monetary regulators can prevent the accumulation of negative
net worth ?
Conclusion
1/2
23


First investigation of NNWs in banking systems of emerging markets
Two exercises: forecasting the size of NNWs when a bank has already
failed and when a bank has not failed yet (sample selection bias)
Propose a new set of identifying hypotheses which may be tested for
both emerging markets and US:
(Н1) Balance sheet falsifications
(Н2) High turnovers on assets
(Н3) Low business margins
(close to affiliated lending and “pocket” banks)
Conclusion
2/2
24

If a bank has already failed, then the NNW will be higher in case of:
1.
2.


Higher turnovers on corporate loans (elasticity=0.40)
Higher reliance on (expensive) retail deposits and higher specialization on granting
(cheap) corporate loans (elasticity=0.35)
3.
Higher bank size (elasticity=0.31)
4.
Higher turnovers on correspondent accounts in the Bank of Russia (elasticity=0.25)
5.
Less capital reported on the eve of license withdrawal (elasticity= –0.25)
6.
Less NPLs reported, but higher LLPs accumulated (elasticity= –0.18)
In-sample forecast: As of 2015M12, 365 out of 697 still operating banks might
face with negative capital. Total estimated NNW = 3.4 – 6.9% of GDP 2015.
Note: after 9 months, 50 out of these 365 banks were failed with total NNW =
0.3% GDP only. Major NNWs are yet to be revealed in the future.
Out-of-sample forecast: 46 banks failed in 2015M6-2015M12 with total NNW =
0.95% of GDP. Our Heckman selection model, estimated up to 2015M6, predicted
38 of that 46 banks with total NNW = 0.93% of GDP