Improving financial distress prediction using textual sentiment of annual reports

Abstract

An accurate prediction of financial distress is beneficial to investors and allows banks and other financial institutions to build an early warning system to avoid risk contagion. This study investigated financial distress prediction using textual sentiment extracted from listed firms’ annual reports in the Chinese market. The sentiments reflected by the firms’ management discussions and analysis (MD&A) sections and audit reports were extracted separately through the application of deep learning algorithms. We found that the sentiment score extracted from MD&A sections was more optimistic compared with that extracted from audit reports. Moreover, the experimental results demonstrated that the modeling performance was significantly improved with the incorporation of textual sentiment scores, and the inclusion of sentiment from audit reports lead to a more significant incremental improvement than that from the MD&A sections. However, when both sentiment scores were included in the modeling input, the improvement in predictive accuracy was insignificant compared to the model using audit report scores only. Our study highlights the predictive power of textual information in annual reports, and shows that the textual sentiment of annual reports should be applied in distress modeling. The results provide implications for the utilization of soft information in credit risk modeling in the context of Chinese market, and such application can be further explored in other areas of operational research studies.

Appendices

Appendix A: Examples sentences with sentiment labels in MD&A sections and audit reports

Segment	Sentiment	Example sentences
MD&A sections	Positive	It is planned to actively extend the intelligent government services business in Shanghai and Guangxi Province
		Sales of social security card chip has experienced substantial growing in 2017 due to the policy adjustment in the short term, and more efforts should be made for the competition in the market of the third-generation social security card chip
	Neutral	We will make an overall plan to connect all sections together hierarchically and horizontally
		We aim to strengthen the cash management and explore more financing opportunities
	Negative	Deferred income has experienced an increase of 15,057,000 yuan due to the delay of the settlement of national scientific funding
		Because the corporate has initiated the procedure of material asset restructuring since 2013, and thus high service fee is incurred
		Iron ore is the main raw materials of steel products, of which the price has strong correlation with national economic structure, economic cycle and industrialisation level, and the demand of iron ore is highly dependent with importation without possession of pricing power which is also impacted by ocean freight
Audit reports	Positive	In our opinion, the financial statements give a true and fair view of the financial position of the corporation at 31 December 2012, and the corporate operations and cash flows in 2012 in accordance with Chinese Financial Reporting Standards
	Neutral	We believe that the audit evidence we have obtained is sufficient and appropriate to provide a basis for our audit opinion
		In making those risk assessments, the auditor considers internal control relevant to the entity’s preparation and true and fair presentation of the financial statements in order to design audit procedures that are appropriate in the circumstances, but not for the purpose of expressing an opinion on the effectiveness of the entity’s internal control
		The property management incomes are the main sources of the firm profit in 2013
	Negative	Until 31 December 2017, the receivables of Shanghai Putian accumulates to 748,956,944 yuan, in that its obligor (Zhejiang Dawei Co., Ltd.) has filed bankruptcy since July 2017
		The corporate has filed an accusation to Liyang Industrial Development Co.,Ltd. due to the dispute in the contract of Tianmu Lake Hotel decoration

Appendix B: Full list of candidate variables

No.	Variables	No.	Variables
1	MDA score	28	Current assets to turnover
2	Audit score	29	Long-term assets to turnover
3	Working capital to total assets	30	Net assets to turnover
4	Retained earnings to total assets	31	Account payable to turnover
5	Earnings before interest and taxes to total assets	32	working capital to turnover
6	Market equity to total liabilities	33	Cash to turnover
7	Total sales to total assets	34	Operating revenue per share
8	Net income to total assets	35	Earnings per share
9	Total liabilities to total assets	36	retention ratio
10	Current ratio	37	Rate of capital accumulation
11	Quick ratio	38	price to earnings
12	Cash ratio	39	dividend to price
13	Owner’s equity to total assets	40	Market value to cash flow
14	Fixed assets to total assets	41	Market value to sales
15	Owner’s equity to fixed assets	42	Net cash flow per share
16	Current liability to total liabilities	43	Net cash flow from operating activities to net profit ratio
17	Long-term liability to total liabilities	44	Net assets per share
18	Owner’s equity to liability	45	Surplus reserves per share
19	Debt to tangible assets ratio	46	Undistributed profits per share
20	Gross operating profit ratio	47	Net cash flow from operating activities per share
21	Net profit to operating revenue	48	Net cash flow from investing activities per share
22	Net profit to current assets	49	Net cash flow from financing activities per share
23	Net profit to fixed assets	50	Current assets to owner’s equity
24	Net profit to owner’s equity	51	Price to book value
25	Return on invested capital	52	Relative size
26	Account receivables to turnover	53	Excess return
27	Inventory to turnover	54	Market volatility