Analysis of trading strategies and asset allocation with deep reinforcement learning

Abstract

In this paper, we examine the empirical performance of investment strategies based on market prices of assets. This project is constructed in a threefold cases. First, we test the performance of technical indicators EMA and MACD by historical price data in United States and South Korea compare with buy and hold strategy. Second, we provide a trading strategy of a single stock with NASDAQ composite index as a trading signal generator. To the end, we investigate Deep Reinforcement Learning for asset allocation portfolio. In each case, we compare the profitability and risk factor with buy and hold or traditional investment strategies. Some strategies show that they beat the market in whole test period, but the others show the different performance depend on the time periods.