Adaptive Algorithmic Trading Using Volatility-Guided Reinforcement Learning: Empirical Analysis in Indian Markets

Abstract

Equity trading markets often experience periods of price and market volatility. These are driven by various factors, and eventually resulting in significant uncertainty for retail investors and widespread financial losses. Volatile markets often lead to frequent trend reversals, diminishing the effectiveness of traditional signal generation methods. Moving average cross signals, or oscillator signal do not ensure a movement in the indicated direction when raised in a volatile environment. Hence it becomes imperative to explore techniques to navigate volatile markets in an algorithmic framework. This paper proposes a reinforcement learning agent which employs volatility as an input, while trying to make profitable high frequency trades. The strategy was backtested on intraday stock data [750 stocks] from multiple indices listed on the National Stock Exchange of India, representing various sections and sectors of the market. It was found to convincingly outperform the buy and hold strategy in bearish and sideways moving conditions, establishing it as a valid scalping option when trying to diversify and hedge portfolios while employing algorithmic trading. The strategy involves the use of Q-Learning as the major decision-driver, optimized by added stochasticity to respond to volatility.