Machine Learning Methods on Fundamentals and Momentum for Long-Term Long/Short Equity Trading Strategies in the High Growth Technology Sector

Abstract

High growth tech companies burn through cash and are often not profitable at the time of going public, and several years after. With negative free cash flows, investors apply a multiple on revenue in order to understand its enterprise value. Enterprise Value-to-Sales multiples have huge volatility and decay, and outside of revenue growth rate, there are no indicators that correlate with the value of that multiple. In this paper, we forecast 1- year forward revenues using machine learning methods of LASSO Regression and Gradient Boosted Trees (XGBoost) on companies within the high growth software space using selected historical fundamental data and daily momentum features, indexed by security and dates. Using the predictions, we analyze the forecasts in order to rank the largest spread in current and predicted revenues. Rather than predicting stock price or market cap growth directly, we predict a fundamental that is critical to valuing the company, thus implicitly projecting future market cap growth. Applied to present trading data, we demonstrate that a portfolio of high growth tech stocks using our selected models on the largest spreads between our 1-year forecasted revenue growth and the current revenue growth outperform benchmarks, and our market-neutral strategies generate diversified, positive returns without taking on excess risk. We present several strategies for prediction and report on selected portfolio methodologies using equal-weight portfolios in order to demonstrate the success of long, short, and market neutral investment strategies on historical trading data.