Enhancing Portfolio Construction with Synthetic Data

Literature proposes a wide array of techniques for portfolio construction, the essence of which relies on the availability of large datasets of the underlying portfolio constituents. The lack of freely available financial data poses a challenging problem for researchers and practitioners. Recently, there has been a rising interest in the use of generative models to augment time-series financial datasets with synthetic data. In this article, the authors examine a deep generative model that can generate high-quality correlation-aware synthetic financial time-series utilizing multi-stock datasets. A holistic assessment of the synthetic data samples is carried out for both distribution characteristics and correlation similarity using visual and quantitative methods. A novel technique is proposed for GAN-assisted risk-based portfolio construction technique that can be incorporated with traditional portfolio management methods. Empirical findings demonstrate that using correlation-aware synthetic data together with real historic market data significantly improves out-of-sample Sharpe Ratio. Utilizing a multi-stock dataset of an emerging market, this technique achieved an improvement in Sharpe Ratio of 18.12% over a benchmark portfolio constructed using only historic data.