مقایسه توانایی الگوریتم یادگیری ماشین آدابوست در تبیین نابهنجاری اقلام تعهدی با استفاده از مدل‌های آربیتراژ، قیمت‌گذاری دارایی‌های سرمایه‌ای و پنج عاملی فاما و فرنچ

تبیین نابهنجاری اقلام تعهدی و جستجوی عوامل ایجاد آن در بازار سرمایه از موضوعات مهم در حوزه مالی است؛ زیرا اثبات وجود نابهنجاری‌های اقلام تعهدی در بازار سرمایه می‌تواند از لحاظ علمی مدل‌های قیمت‌گذاری سهام را به چالش کشیده و نقش عواملی غیر از ریسک سیستماتیک در پیش‌بینی بازده در بازار را برجسته کند؛ از اینرو هدف اصلی این پژوهش، مقایسه توانایی الگوریتم یادگیری ماشین آدابوست در تبیین نابهنجاری اقلام تعهدی با استفاده از مدل‌های آربیتراژ، قیمت‌گذاری دارایی‌های سرمایه‌ای و پنج عاملی فاما و فرنچ است. برای دستیابی به هدف پژوهش، نمونه‌ای متشکل از 120 شرکت پذیرفته ‌شده در بورس اوراق تهران طی دوره زمانی 1387-1398 با استفاده از الگوریتم یادگیری ماشین آدابوست بررسی شده است. نتایج پژوهش نشان داد اگر تاثیر نابهنجاری اقلام تعهدی بر بازده سهام در نظر گرفته شود، ریسک‌های حاصل از مدل‌های آربیتراژ، قیمت‌گذاری دارایی‌های سرمایه‌ای و مدل پنج عاملی فاما و فرنچ کاهش پیدا خواهد کرد و در نتیجه به قیمت واقعی سهام نزدیک‌تر خواهد بود که این امر باعث افزایش اعتماد سرمایه‌گذاران می‌شود؛ لذا اضافه شدن نابهنجاری اقلام تعهدی در مدل‌های مالی آربیتراژ، قیمت‌گذاری دارایی‌های سرمایه‌ای و مدل پنج عاملی فاما و فرنچ، منجر به بهبود در ارزیابی بازده سهام می‌شود و نابهنجاری اقلام تعهدی در بورس اوراق بهادار تهران وجود دارد. نتایج نشان داد نابهنجاری اقلام تعهدی توسط مدل پنج عاملی فاما و فرنج در ارزیابی بازده سهام بهتر توضیح داده می‌شود. به بیانی دیگر، توان پیش‌بینی مدل پنج عاملی فاما و فرنچ نسبت به مدل آربیتراژ و قیمت‌گذاری دارایی‌های سرمایه‌ای بیشتر است.

Comparison of the ability of AdaBoost machine learning algorithm to explain the accrual anomaly using arbitrage pricing model, capital asset pricing model and FamaFrench fivefactor model

1 IntroductionPurpose of research is investigating the effect of accrual anomaly on stock return short arbitrage financial model and capital asset pricing by using a neural network. Bankruptcy of companies is one of the ways that leads to wasting resources and not taking advantage of investment opportunities. Predicting financial distress can alert companies to the occurrence of financial distress and subsequent bankruptcy with the necessary warnings so that they can take appropriate action according to these warnings and investors can take advantage of unfavorable opportunities. Recognize and invest their resources in the right opportunities and places. One way to predict the continuity of corporate activity is to use models to predict financial distress; Therefore, the main purpose of this study is to predict the financial distress of companies based on working capital management using artificial neural network. 2 Research questionsConsidering that no coherent research has been done in the field of forecasting financial distress of companies based on working capital management using neural networks method, this research can be an introduction to identify the impact of the role of capital management in Circulation, in order to find solutions to increase the continuity of the company; Therefore, the main questions that this study seeks to answer are as follows. What is the accuracy of predicting companies’ financial distress using artificial neural network method based on working capital management variable? How accurate are artificial neural network models, decision tree, support vector machine, multiple audit analysis, and logistic regression in predicting corporate financial distress? 3 Methods In order to achieve the purpose of the research, samples consisting of 120 companies listed on the Tehran Stock Exchange during the period 20082019 have been studied. In this study, the hypotheses have been tested using the AdaBoost machine learning algorithm, and sales arbitrage pricing models, capital asset pricing model, and FamaFrench fivefactor model have been used to analyze the anomalies of accruals. 4 ResultsIn this research after the testing of research hypotheses we got this result, if the effect of accrual anomaly on stock returns is considered, the risks of arbitrage pricing models, capital asset pricing model and the Fama and French fivefactor model will be reduced and thus closer to the real stock price. This will increase the return and confidence of investors. The results of comparing the three models based on AdaBoost machine learning algorithm showed that the development of FamaFrench fivefactor model reduces the neural network training error with AdaBoost algorithm to a greater extent than arbitrage pricing models and capital asset pricing model. In explaining the abnormality of items, it has an obligation on stock returns. This result shows the effectiveness of the inclusion of accruals in the securities pricing models. In other words, the addition of anomalies of accruals to arbitrage financial models, capital asset pricing model, and the FamaFrench fivefactor model leads to an improvement in stock returns. 5 Discussion and ConclusionInvestors should distinguish between the stability of profit components (cash and accrual) when valuing companies. The disregard of this difference has made investors optimistic about the future performance of companies when the FirmSpecific Discretionary Accruals is high, and pessimistic about the future of companies when FirmSpecific Discretionary Accruals is low. So, the purpose of testing the first research hypothesis is to investigate the effect of adding accruals anomalies to stock returns of the financial arbitrage sales model; Therefore, using the Adabost machine learning algorithm, the expected return and the actual return have been calculated to determine its impact on market indicators. For this purpose, calculations have been performed without the anomalous effect of accruals. The test results of the first hypothesis showed that the percentage of accuracy and prediction of expected return has multiple errors. Then, calculations were performed based on the effect of accrual anomalies on market indices. The result of these calculations showed a reduction in errors in expected returns. The results of the second hypothesis showed that the addition of anomalies of accruals to the capital asset pricing model in assessing stock returns increases the predictive power of the model. This finding indicates that accruals have informational value. And plays an important role in the stock price valuation process; Because it reduces the scheduling problems and the lack of conformity in the cash figures. The purpose of testing the third hypothesis of the research is to investigate the effect of adding accruals anomalies to the stock returns of the Fama and French fivefactor model. The results of testing the third hypothesis showed that adding anomalies of accruals to the capital asset pricing model in assessing stock returns increases the predictive power of the model. In general, the results of the third hypothesis of the research indicate that investors in the processing of accounting information, especially accruals and consequently the valuation of corporate stocks, are faced with incorrect pricing. The purpose of testing the fourth hypothesis of the research is the ability of the fivefactor model of Fama and French compared to the traditional model of arbitrage of sales and pricing of capital assets in explaining the anomalies of accruals. The results of the fourth hypothesis test showed that there is an anomaly in accruals on the Tehran Stock Exchange and this anomaly is better explained by the fivefactor model of Fama and Farang. For example, the real rate of return in 1387 is equal to 29.103, the projected rate of return after adding the anomaly of accruals to the arbitrage model of 021/28; The projected rate of return after adding the anomaly of accruals to the capital asset pricing model is 25.471; If the projected rate of return after adding the anomaly of accruals to the five factors of Fama and French is equal to 30.221. This trend continues in the same way for the rest of the years under study, and indicates that the predictability of the Fama and French fivefactor model is greater than the arbitrage model and pricing of capital assets.  Keywords: Accrual Anomaly, Arbitrage Pricing Model, Capital Assets Pricing Model, FamaFrench FiveFactor Model.