Two stage portfolio optimisation trough sentiment analysis: a data envelopment analysis based approach

The existing portfolio optimisation studies focus on the portfolio selection problem as a pure optimisation problem, and they mostly ignore the importance of the asset selection. However, in actual investment process, asset selection is equally important and the returns of portfolios created according to the different asset selection have significant differences. As a solution to this problem, we propose two-stage investment strategy to construct an effective portfolio consisting of BIST30 stocks. In the first stage, we identify investment worthy stocks by establishing an asset selection schema with variable return to scale data envelopment analysis (VRS-DEA) using stock historical data, correlation coefficients and investor sentiment data. We used Google Trend and AR popularity index data as proxy for investor sentiment. Furthermore, we used two different techniques to ensure low correlation between selected stocks when performing portfolios. In both techniques, our empirical results show that the DEA-based stock selection with investor sentiment data can improve the out-of-sample performance of different investment strategies and our results are supportive for under-diversification theory. Lastly, we used first stochastic dominance (FSD) test as a robustness test to verify our conclusions and it is clear that FSD test results are consistent with our conclusions.