Prediction of short-term stock price response to news

We tested a lot of machine learning methods in Table 4. According to prediction's statistics we see that weightier models have significant improvements. In fact, simple models have weak predictive ability, this means that models have limited hypothesis space and they couldn't be better than they are. As for standard simple linear regressions we must to point that their performance is lower than standard portfolio's theory models. However, standard machine learning models are better than simple regressions and portfolio's models. The decision's logic is more complex and involved in these models. Moreover, standard machine learning models are newer than previous models. Deep learning models are the best in this way, because they have spanning windows, back looking, forward vision and have more inside parameters. Big difference in statistics starts between small ANN, which has better performing than previous nine models. We included dropout technique in next models and added some reliable methods for performance. Final model of the research is Bidirectional Long-Short Memory Neural Network with custom losses, optimizer and accuracy metrics. We achieved such a result with nesterov's optimization and magnitude in stochastic gradient descent.

Summarizing results, we have a new complex model for creating forecasting after summarizing all models in general. Statistics of models and components demonstrate impact of different models. Comparison between simple event-study approaches and modified approaches gave us more relevant general model, which has more reliable description statistics than standard methods.

USDX and LIBOR were included in general model. They also have impact on model's relevant, but not such huge as the sentiment analysis. In any case it has developed general model.

Sentiment analysis demonstrates a significant impact in research, because it isn't a standard additional method for stock forecasting, moreover, it increases rating metrics for general model. It has bias in predictions on X_company output, because every company has own correlations, reactions and local market efficiency. On the one hand it gave good results and better model than standard approach, on the other hand it needed deep comprehension of programming skills to parse and include all variables and transformations to model.

As for hypotheses, we can construct model with volatility outputs or 3-signal outputs, it depends on tasks. Model can predict both of types of outputs. We accept the first hypothesis. NLP or sentiment index can improve models based on historical data. We proved it with different models and descriptions statistics, also sentiment index is significant in OLS estimation. As for the second hypothesis, we accept it for some reasons:

1. A type of final machine learning wrapper is important for predictions

2. Different methods have different results, but this is due to more complex models or State-of-Art models, which have trained only on GPU by reason of their weight

3. Fine-tunning is the most important part of any model. The critical task in modeling isn't labeling or featuring, the task is right hyperparameters, right batch sizes, epochs, dropouts and other parameters of models

Summarizing all things, we can answer the main question of research about prediction power of markets. Non-standard approaches can beat only historical models and they are more reliable in predictions.

Conclusion

Our investments in the researcher's area of stocks are significant. We have demonstrated that sentiment analysis is an important part of future research in this area. The news background of exchanges has a direct effect on the price of stocks. The BERT model has helped us to improve existing models of portfolio theory and has shown that they are becoming more reliable in research. Future researchers in this field can begin their research by parsing news data and studying BERT technology. All these advantages will help them to discover something new in market charts.

Despite this fact, we found that the wrapper of the general model is very important for accurate predictions. It is not possible to achieve a serious result without knowledge in machine learning. The baseline study began with exquisite data on time series and transforming them into a random walk. After that, you have a stationary series that can be used in further research and apply various models. According to tables, any researcher could select interesting model and it would be a base model. His tasks would be finding new variables, testing new hypotheses and constructing more complex models than we had.

On the other hand, our research hasn't volatility predictions, «up» or «down» predictions and back testing strategy with portfolio. All of these things are another type of research or future work. Despite of that, we have significant description statistics and lagged predictions with LSTM. In other words, our final model repeats a real smoothed market graph and the next step in this situation is trading's strategy or simulation.

The contribution of this study is that it is efficient to reduce the prediction error by using a combination of previous researcher's models and new model, which contains modified wrapper and sentiment index, from the same data instead of using these models separately. As for measurable research's values, we confirmed main advantages of Theoretical background. Authors demonstrated that simple model couldn't be reliable in nowadays and we also demonstrated. We have found from this study that integrating event information with a prediction model plays very important roles for forecasting more accurately. Above all, we described about simple regressions with linear relationship between variables. Traditional statistical models are widely used in economics for time series predictions. We claimed that NNs substantially outperform traditional statistical methods.

We have done a lot of methods and all of methods are suitable for approach. We showed important techniques of different areas to construct new general model and all of these techniques would be reliable in future. We demonstrated custom fine-tuning for better results, hence, Fine-tuning is one of critical parts of our research, however, at the begging of work we didn't understand this problem. The text information in the stock market such as news is not fully utilized by us. There is a way to further improve the performance of our proposed model. The next step of this research is comparing different NLP models and fine-tuning their hyper parameters.

References

1. Banz Rolf W., 1981, The relationship between return and market value of commonstocks, Journal of Financial Economics 9, 3-18.

2. Chan K.C., Nai-fu Chen, 1991, Structural and return characteristics of large and small firms, Journal of Finance 46, 1467-1484.

3. Campbell, John Y., Andrew W. Lo., A. Craig MacKinlay, 1997, The Econometrics of Financial Markets, Princeton University Press, Princeton, New Jersey, USA.

4. Daniel, Kent, and Sheridan Titman, 1997, Evidence on the characteristics of cross-sectional variation in stock returns, Journal of Finance 52, 1-33.

5. Davis, James L., Eugene F. Fama and Kenneth R. French, 2000, Characteristics, covariance and average returns: 1929 to 1997, Journal of Finance 55, 389-406.

6. DeBondt, Werner F.M. and Richard H. Thaler, 1985, Does the stock market over react, Journal of Finance 40, 793-805.

7. Fama Eugene F., Kenneth R. French, 1992, The cross-section of expected stock returns, Journal of Finance 47, 427-465.

8. Fama Eugene F., Kenneth R. French, 1992, The cross-section of expected stock returns, Journal of Finance 47, 427-465.

9. Fama Eugene F., Kenneth R. French, 1993, Common risk factors in the returns on stocks and bonds, Journal of Financial Economics 33, 3-56.

10. Fama Eugene F., Kenneth R. French, 1995, Size and book-to-market factors in earnings and returns, Journal of Finance 50, 131-155.

11. Fama Eugene F., Kenneth R. French, 1996, Multifactor explanations of asset pricing anomalies, Journal of Finance 51, 55-84.

12. Gibbons Michael R., Stephen A. Ross, Jay Shanken, 1989, A test of the efficiency of a given portfolio, Econometrica 57, 1121-1152.

13. Keim Donald, 1983, Size-related anomalies and stock return seasonality: further empirical evidence, Journal of Financial Economics 12, 13-32.

14. Kothari S.P., Jay Shanken, Richard G. Sloan, 1995, Another look at the cross-section of expected stock returns, Journal of Finance 50, 185-224.

15. Lakonishok, Josef, Andrei Schleifer and Robert W. Vishny, 1994, Contrarian investment, extrapolation and risk, Journal of Finance 49, 1541-1578.

16. MacKinlay A. Craig, 1995, Multifactor models do not explain deviations from the CAPM, Journal of Financial Economics 38, 3-28.

17. Muneesh Kumar and Sanjay Sehgal, 2000, Company characteristics and common stock return: the India experience, working paper, University of Delhi.

18. Sehgal, Sanjay, 2001, Investor behavior in Indian capital markets, working paper, University of Delhi.

19. A.V. Devadoss, T.A.A. Ligori, “Forecasting of stock prices using multi layer perceptron,” Int J Comput Algorithm, vol. 2, pp. 440-449,2013.

20. J. G. De Gooijer and R. J. Hyndman, “25 years of time series forecasting,” International journal of forecasting, vol. 22, no. 3, pp. 443-473, 2006.

21. V.K. Menon, N.C. Vasireddy, S.A. Jami, V.T. N. Pedamallu,

22. V. Sureshkumar, and K. Soman, “Bulk price forecasting using spark over data set,” in International Conference on Data Mining and Big Data. Springer, 2016, pp. 137-146.

23. G.E. Box, G.M. Jenkins, G.C. Reinsel, and G. M. Ljung, Time series analysis: forecasting and control. John Wiley & Sons, 2015.

24. G. Batres-Estrada, “Deep learning for multivariate financial time series,” ser. Technical Report, Stockholm, May 2015. P. Abinaya, V.S. Kumar, P. Balasubramanian, V.K. Menon, “Measuring stock price and trading volume causality among nifty50 stocks: The toda yamamoto method,” in Advances in Computing, Communications and Informatics (ICACCI), 2016 International Conference IEEE, 2016, pp. 1886-1890.

25. J. Heaton, N. Polson, J. Witte, “Deep learning in finance,” arXiv preprint arXiv:1602.06561, 2016.

26. H. Jia, “Investigation into the effectiveness of long short term memory networks for stock price prediction,” arXiv preprint arXiv:1603.07893, 2016.

27. Y. Bengio, I.J. Goodfellow, A. Courville, “Deep learning,” Nature, vol. 521, pp. 436-444, 2015.

28. H. White, Economic Prediction Using Neural Networks: The Case of IBM Daily Stock Returns, ser. Discussion paper- Department of Economics University of California San Diego. Department of Economics, University of California, 1988.

29. B.G. Malkiel, “Efficient market hypothesis,” The New Palgrave: Finance. Norton, New York, pp. 127-134, 1989.

30. X. Ding, Y. Zhang, T. Liu, J. Duan, “Deep learning for event-driven stock prediction.” in IJCAI, 2015, pp. 2327-2333.

31. J. Roman and A. Jameel, “Backpropagation and recurrent neural networks in financial analysis of multiple stock market returns,” in System Sciences, 1996., Proceedings of the Twenty-Ninth Hawaii International Conference on,, vol. 2. IEEE, 1996, pp. 454-460.

32. M.-C. Chan, C.-C. Wong, C.-C. Lam, “Financial time series forecasting by neural network using conjugate gradient learning algorithm and multiple linear regression weight initialization,” in Computing in Economics and Finance, vol. 61, 2000.

33. J. Roman, A. Jameel, “Backpropagation and recurrent neural networks in financial analysis of multiple stock market returns,” in System Sciences, 1996., Proceedings of the Twenty-Ninth Hawaii International Conference on,, vol. 2. IEEE, 1996, pp. 454-460.

34. E.W. Saad, D.V. Prokhorov, D.C. Wunsch, “Comparative study of stock trend prediction using time delay, recurrent and probabilistic neural networks,” IEEE Transactions on neural networks, vol. 9, no. 6, pp. 1456-1470, 1998.

35. O. Hegazy, O.S. Soliman, M.A. Salam, “A machine learning model for stock market prediction,” arXiv preprint arXiv:1402.7351, 2014.

36. K.-j. Kim, I. Han, “Genetic algorithms approach to feature discretization in artificial neural networks for the prediction of stock price index,” Expert systems with Applications, vol. 19, no. 2, pp. 125-132, 2000.

37. Y. Kishikawa and S. Tokinaga, “Prediction of stock trends by using the wavelet transform and the multi-stage fuzzy inference system optimized by the ga,” IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences, vol. 83, no. 2, pp. 357-366, 2000.

38. S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural computation, vol. 9, no. 8, pp. 1735-1780, 1997.

39. Murtaza R., Harshal P., Shraddha V.: Predicting Stock Prices Using LSTM, International Journal of Science and Research (IJSR) Volume 6 Issue 4, 2017.

40. LeCun Y., Bottou L., Bengio Y., Haffner P.: Gradient-based learning applied to document recognition. Proceedings of the IEEE 86(11), 2278-2324 (1998)

41. Bouvrie J.: Notes on Convolutional Neural Networks[J]. Neural Nets, 2006.

42. Hubel D.H., Wiesel T.N.: Receptive fields and functional architecture of monkey striate cortex. The Journal of Physiology 195(1), 215-243 (1968)

43. Murtagh F., Starck J.L., Renaud O.: On neuro-wavelet modeling [J]. Decision Support Systems, 2004, 37(4):475-484

Размещено на Allbest.ru