Over the past twenty years, trading in financial markets has evolved from a human-oriented process to one that is highly automated. One of the most influential and revolutionary processes in financial markets is algorithmic trading. The focus of this work is to increase trading efficiency by improving the accuracy of intraday volume forecasts, which are used in algorithmic trading. An intraday volume forecast predicts the distribution of trading volume throughout the day, and allows traders to make better decisions regarding the timing and quantity of their trades. An accurate intraday volume forecast is an important input to trading decisions and will ultimately improve traders' ability to meet benchmarks, such as volume weighted average price. This work seeks to understand the performance of three classes of models: moving average, exponentially weighted moving average, and average exponentially weighted moving average, for intra-day volume prediction over a large sample of U.S. equities. For each model, we explore a broad range of parameterizations and seek to understand how various factors affect model performance. Models are evaluated based on a variety of different metrics, such as mean square error and maximum absolute deviation. We report on the best performing models for a variety of stocks and scenarios. Overall, the average exponentially weighted moving average model performed the best across the examined parameters.
展开▼
