Abstract:
هدف: پژوهش حاضر با شناسایی عوامل موثر بر نقدشوندگی، مدلی برای پیشبینی وضعیت نقدشوندگی سهام، در بازار اوراق بهادار تهران ارائه کرده است.
روش: 23 عامل مشخصشده در مطالعات کتابخانهای، بر اساس دادههای 154 شرکت فعال در فاصله زمانی 1388 تا 1398 استخراج و در قالب دو خوشه افراز شد. با استفاده از معادلات ساختاری با رویکرد حداقل مربعات جزئی، اعتبار متغیرهای شناساییشده با معیار حاصل از خوشهبندی ارزیابی شد.
یافتهها: ارزیابی ارتباط متغیرها در مدلهای یادگیری ماشینی نشان داد که جریان نقدی غیرعادی، هزینه اختیاری غیرعادی، خطای تخمین اقلام تعهدی، تفاوت بین سرمایه در گردش تحقق یافته و موردانتظار، سهام شناور آزاد، دوره تصدی حسابرس، حقالزحمه حسابرسی، سهم بازار حسابرس، محافظهکاری و تغییر حسابرس، در خوشهبندی بیشترین تاثیر را دارند. سرانجام بهترین مدل یادگیری ماشینی، بر اساس آموزش و آزمون انتخاب شد.
نتیجهگیری: نتایج نشان میدهد که متغیرهای مستقل، بیش از 72 درصد از تغییرات نقدشوندگی را توضیح میدهند. همچنین مدل شبکههای عصبی در مقایسه با سایر مدلهای یادگیری ماشینی، توان پیشبینی بیشتری دارد و با 32/99 درصد صحت برازش، مناسبترین مدل پیشبینی نقدشوندگی است.
Objective: In this study, a market liquidity prediction model is proposed for the Tehran Stock Exchange (TSE) by identifying the factors affecting liquidity. Methods: Based on the data of 154 Tehran Stock Exchange (TSE)-listed companies for the 2009–2020 period, the values of 23 factors were extracted and divided into two clusters. The partial least squares structural equation modeling (PLS-SEM) technique was employed to validate the variables extracted through the evaluation criterion and to determine their power to explain changes. Results: The relationships among the variables were evaluated by machine learning models. The results indicated the 10 variables of abnormal cash flow, abnormal discretionary expenses, accrual estimation error, the difference between fulfilled and expected working capitals, free float, auditor’s tenure, audit fees, auditor’s market share, conservatism, and change of auditor to have the greatest effect on clustering. Finally, the best machine learning model was selected through training and testing. Applying the logistic regression model showed that for the calculated value of a dependent variable greater than or equal to 3.391, the illiquidity of the company’s shares is definite, and for the calculated value of the dependent variable less than 3/391, the liquidity of shares of the company is definite. Finally, the best machine learning model was selected through training and testing. Conclusion: Based on library studies, 89 different liquidity criteria have been used around the world in the studies on the subject of liquidity and various classifications, including measures with more or less frequency and one-dimensional or multi-dimensional measures. There is less agreement on the best measure and the correlation among most of these measures indicating that the use of an inappropriate measure may lead to incorrect conclusions. Using Minitab software, the dependent variable was compared with 5 commonly used criteria (measures) known in library studies (Liu measure, Amihud illiquidity measure, zero measure, number of trading day measure, and trading volume measure). The results confirm the statistically significant difference between the widely used and newly obtained measures. It should be noted that the dependent variable was extracted using non-supervisory patterns in machine learning software and a statistically significant difference between the widely used criteria and the obtained variable was proved. Thus, the resulting variable is a criterion for liquidity. According to the results, the independent variables explain more than 72 percent of the changes in liquidity. Moreover, the neural network model was more capable of prediction than the other machine learning models. In fact, it was proved as the best liquidity prediction model with 99.32 percent of the fitness accuracy.