Failure prediction of Indian Banks using SMOTE, Lasso regression, bagging and boosting

Figures & data

Table 1. The information about the independent and dependent variables used in the analysis are given below

NAME	TYPE	DEFINITION
Status	Categorical	Binary indicator equivalent to 1 for the failed banks in the year immediately prior to the failure event 0 in all other years.
Total Assets	Quantitative	Cash and assets due to banks, total earning assets, foreclosed real estate, fixed assets, and other assets.
Equity	Quantitative	Common equity, non-controlling interest, securities revaluation reserves and foreign exchange revaluation reserves.
Total Liabilities	Quantitative	Total assets minus equity
Total Provision	Quantitative	Net loans minus reserves for impaired loans.
Deposits	Quantitative	Customer deposits, bank deposits, other deposits, and short-term borrowings.
Profit after tax	Quantitative	Profit after tax is the net profit earned by the company after reducing all expenses.
Total Capital	Quantitative	Total capital = total liabilities plus equity
Reserves and Funds	Quantitative	The reserve fund is a savings account or other highly liquid asset set apart by banks to meet any future costs or financial obligations
Return on assets	Quantitative	Net Income/Average Total Assets
Net Income	Quantitative	Post-tax profit
Net Interest Revenue	Quantitative	Gross interest and dividend income minus total interest expense
Other Operating Income	Quantitative	Any other sustainable income which is related to the company’s core business.
Overheads	Quantitative	Personnel and other operating costs.
Z-score	Quantitative	A measure is inversely related to the probability of the bank’s insolvency.
Loan Loss Reserves/Loans	Quantitative	It signifies how much funds have been put apart for potential losses.
Equity/Assets	Quantitative	Evaluates the amount of security the bank enjoys by its equity
Equity/Net Loans	Quantitative	Measures the equity insulation available to take up losses on the loan manuscript
Equity/Deposits	Quantitative	Estimates the amount of everlasting funding relative to undersized funding.
Equity/Liabilities	Quantitative	Also identified as the capitalization ratio and it is the inverse of the leverage ratio.
Net Interest Margin	Quantitative	Net interest income articulated as a percentage of earning assets.
Cost/Income	Quantitative	Estimates the costs of managing the bank, the main element of salaries, as a proportion of income produced before provisions.
Net Loans/Assets	Quantitative	Reveals what proportion of the resources of the bank are coupled up in loans
GDP Growth	Quantitative	The development rate of inflation rectified GDP.
Inflation	Quantitative	Logarithmic change of the GDP deflator on year to year basis
C3/All	Quantitative	Bank concentration from Top 3 banks
C5/All	Quantitative	Bank concentration from Top 5 banks

Figure 1. SMOTE algorithm (Synthetic data generation method)

Figure 2. Ensemble learning method

Figure 3. Bagging procedure for algorithm learning

Table 2. Descriptive statistics for private and public sector Indian banks for the period of 2000–2017

NAME	Minimum	Median	Mean	Maximum	Standard Deviation
Total Assets	0.00	472726.00	1184784.00	27059663.00	2238478.00
Equity	−9931.00	26865.00	78229.00	1882861.00	160046.20
Total Liabilities	866.00	401609.00	952141.00	20447514.00	1715036.00
Deposits	866.00	401609.00	952140.00	20447514.00	1725551.00
Profit after tax	−60892.10	3349.90	8269.00	145496.40	19065.69
Total Capital	0.00	472726.00	1184516.00	27059663.00	2238607.00
Reserves and Funds	−34971.00	23374.00	73863.00	1874887.00	158809.10
Return on assets	−6.50	0.90	0.85	4.46	0.81
Net Income	0.00	54039.00	132996.00	2700874.00	245053.00
Net Interest Revenue	−14064.00	11839.00	29703.00	625481.00	58665.92
Other Operating Income	79.50	40600.30	103652.60	2075392.80	187286.50
Growth Overheads	34.30	23469.50	61682.30	1139568.90	105937.90
Z-score	−3.27	2.00	2.29	11.46	2.05
Loan Loss Reserves/Loans	0.00	0.02	0.04	0.52	0.04
Equity/Assets	−50.60	0.06	0.01	0.95	1.77
Equity/Net Loans	−0.12	0.11	0.15	11.43	0.45
Equity/Deposits	−0.06	0.07	0.10	11.68	0.43
Equity/Liabilities	−0.98	0.06	0.11	19.86	0.77
Net Interest Margin	0.00	0.03	0.03	0.68	0.04
Cost/Income	0.92	1.57	1.64	22.75	0.78
Net Loans/Assets	0.00	0.57	0.54	0.74	0.11
GDP growth	0.00	0.13	0.13	0.20	0.04
Inflation CPI	2.23	6.32	6.91	14.97	3.23
C3/All	0.00	0.28	0.25	0.32	0.11
C5/All	0.00	0.38	0.33	0.41	0.14

Table 3. Preliminary data analysis the Table 3 describes the descriptive statistics of accounting profiles to survived vs. Failed banks. The t-tests on the mean difference for different classifications of banks are given below

	I	II
Variables	Survive	Fail
Profit After Tax	10017.56	2066***
Total Assets	1435595	300492**
Return on Net worth	0.92	0.58***
Equity	95788	15946***
Total Liabilities	1339806.8	284545.6***
Total Provision	35391.59	6797***
Loans	935094	246853***
Net Interest Revenue	35929	7686.814 ***
Other operating income	125046	27887***
Growth overheads	74247	17252***
Loan Loss Reserves/Loans	0.03	0.04
Equity/Assets	−0.01	0.06
Equity/Net loans	0.13	0.21
Equity/Deposits	0.09	0.14
Equity/Liabilities	0.1	0.12
Net Loans/Assets	0.55	0.49
Net Interest Margin	0.03	0.03
Cost/Income	1.61	1.72
Z-score	2.34	2.13**
Inflation CPI	6.98	6.6
C3 All	0.26	.22***
C5 All	0.25	0.19***
GDP growth	0.13	0.12

(***, **,* implies significance at the 1%, 5% and 10% respectively)

Table 4. Confusion matrix—Actual values vs predicted results

Predicted Results		Actual Values
		PositiveNegative
	Positive	True Positive	False Positive (Type-I error)
	Negative	False Negative (Type-II error) Sensitivity	True Negative (Specificity)

Table 5. Actual vs predicted—logistic regression

		Actual
Predicted		1	0
	1	41	5
	0	74	132

For the random forest, accuracy and Type-II error are 71.8% and 58.26% respectively.

Table 6. Actual vs predicted—random forest method

		Actual
Predicted		1	0
	1	48	4
	0	67	133

For AdaBoost, the accuracy and Type-II error is 98.8% and 1.73% respectively.

Table 7. Actual vs predicted—AdaBoost method

		Actual
Predicted		1	0
	1	113	1
	0	2	136

Table 8. Accuracy and Type-II error of the different algorithms on the test data

Techniques	Accuracy	Type-II error	Accuracy/Type-II error
Logistic Regression	68.65%	64.34 %	1.066
Random Forest	71.8%	58.26 %	1.23
AdaBoost	98.8%	1.73%	57.10