Online Portfolio Momentum: Exponential Gradient and Follow the Leader
Summary
This overview compares online portfolio momentum approaches across six equity and market-index datasets. Exponential Gradient updates portfolio weights using recent relative performance, with a learning rate and regularization intended to limit abrupt changes. Follow the Leader selects weights based on the best-performing constant rebalanced portfolio over the observed history, while Follow the Regularized Leader adds a penalty to discourage rapid weight changes. The article describes several Exponential Gradient update variants and examines how parameter choices affect allocation behavior.
The historical examples show that results depend strongly on market conditions. Momentum approaches can benefit when an asset sustains a strong trend, as illustrated by Finland’s rise in the developed-market index sample and Amazon’s gains in the recent US equity sample. They fare poorly in declining or choppy periods, and short-sale restrictions limit their ability to respond to broad declines. The reported comparisons are retrospective and involve parameter tuning on historical data, so they do not establish future profitability. Concentration in an exceptional winner can also make performance volatile and vulnerable to a trend reversal.
Key ideas
- Exponential Gradient adjusts portfolio weights using recent returns while regularization moderates the size of changes.
- Follow the Leader uses the observed history to estimate weights for the best constant rebalanced portfolio.
- Regularization can reduce rapid portfolio shifts but may also limit how closely a strategy follows recent winners.
- Momentum performed best in the examples when a strong trend persisted and struggled in falling or erratic markets.
- The historical results depend on the chosen dataset and tuned parameters, and concentrated positions can increase volatility.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.