Naslov
TIME INTERVAL CHOICES IN THE PERFORMANCE ESTIMATION OF FORECASTING DIRECTION OF STOCK PRICE CHANGES

Autori
Vlah Jerić, Silvija

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, prošireni sažetak, znanstveni

Izvornik
BOOK OF ABSTRACTS / - , 2022, 71-72

Skup
International Scientific Conference Technology, Innovation and Stability: New Directions in Finance

Mjesto i datum
Zagreb, Hrvatska, 5-6.05.2022

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Recenziran

Ključne riječi
Technical trading ; stock price prediction ; financial forecasting ; classification algorithms ; machine learning
(Tehničko trgovanje ; predviđanje cijena dionica ; financijsko predviđanje ; algoritmi klasifikacije ; strojno učenje)

Sažetak
An important task in every machine learning project is assessing the performance of the selected method. However, the choice of performance estimation procedure for this task can affect the selection since the performance estimation can substantially differ depending on the procedure used. In time series forecasting tasks there is also the question of time, i.e. choosing the evaluation method includes also decisions regarding time intervals choices, such as window length, but also some other time related peculiarities. A well-known procedure for performance estimation based upon data splitting for cross validation that basically moves the training and test sets in time is known as rolling forecasting origin technique and can be used for tuning the algorithms and selecting the optimal model, as well as for evaluating its’ performance on new data. There are different forms of this technique and all include choices on time interval length, besides others. The main objective of this work is to investigate the impact of time related choices within performance estimation procedures for forecasting the direction of stock price changes using technical indicators. The directions of future price movements are predicted by classifying them into two classes, depending on whether the closing stock price went up or down for a given horizon. Several different machine learning algorithms are evaluated during this process in order to control for differences in algorithms capabilities. Ten indices from CEE (Central and Eastern European) and SEE (Southern and Eastern European) countries are chosen for research in an attempt to investigate their behaviour in the light of the behaviour of bigger and more researched markets. Besides time intervals underlying the process of estimation of predictive performance of system, the predictive system performance will depend on forecasting horizon, time intervals for computing the features, for tuning machine learning algorithms and possible other time related choices. This analysis also aims at investigating how varying the forecast horizon and the input window length for calculating technical indicators in conjunction with time interval choices within the performance estimation procedure affects performance estimation of different machine learning algorithms on forecasting the direction of change of chosen stock market indices. Namely, in a previous study it was investigated how varying the forecast horizon and the input window length for calculating technical indicators affect the predictive performance on their own. In respect to similar research conducted on e.g. S&P 500 Index stocks, that previous analysis on CEE and SEE countries did not find exactly the same pattern of highest system performance for each forecast horizon value when the input window length is approximately equal to the horizon. Instead, for some algorithms the forecasts seemed to be better for short horizons in general. However, it was still better when the input window length was approximately equal to the horizon. Also, there was a notable difference in the performance change with the increase of forecasting horizon where some algorithms performed very well for short horizons and then deteriorated substantially as the forecasting horizon increased, while others seemed to have more consistent performance for different horizons. In this newest study, the analysis is extended to include time related choices within the performance evaluation procedures themselves.

Izvorni jezik
Engleski

Znanstvena područja
Interdisciplinarne tehničke znanosti, Ekonomija, Interdisciplinarne društvene znanosti

POVEZANOST RADA