Big data–enabled sign prediction for Borsa Istanbul intraday equity prices

Abstract

This paper employs a big data source, the Borsa Istanbul's “data analytics” information, to predict 5-min up, down, and steady signs drawn from closing price changes. Seven machine learning algorithms are compared with 2018 data for the entire year. Success levels for each method are reported for 26 liquid stocks in terms of macro-averaged F-measures. For the 5-min lagged data, nine equities are found to be statistically predictable. For lagged data over longer periods, equities remain predictable, decreasing gradually to zero as the markets absorb the data over time. Furthermore, economic gains for the nine equities are analyzed with algorithms where short selling is allowed or not allowed depending on these predictions. Four equities are found to yield more economic gains via machine learning–supported trading strategies than the equities' own price performances. Under the “efficient market hypothesis,” the results imply a lack of “semistrong-form efficiency.”