ARTFIMA Forecasting and a Comparison of Time-Series Trading Strategies

Summary

The article explains how ARTFIMA extends ARFIMA with a tempering parameter that controls how quickly long-memory effects decay, allowing a model to represent both short- and long-term dependence. It outlines the roles of the autoregressive and moving-average terms, fractional integration, and tempering, then describes estimating an ARTFIMA model with Whittle estimation in R. The practical section builds an event-driven strategy that compares forecasts from ARIMA, ARFIMA, and ARTFIMA models on Apple data.

Reported results show ARIMA with the highest annual return among the model strategies and the best Sharpe ratio, while ARFIMA and ARTFIMA have lower annual volatility. The comparison is limited: it uses one stock and a stated sample, and omits commissions, slippage, risk management, and optimization of the estimation span. The article also suggests using model forecasts as features for machine learning, but does not test that extension.

Key ideas

• ARTFIMA adds a tempering parameter to fractional integration to model the decay of long-memory effects.
• The article describes Whittle estimation of an ARTFIMA model in R.
• Its backtest compares ARIMA, ARFIMA, and ARTFIMA forecast signals on Apple returns.
• The reported ARIMA strategy has the strongest return and Sharpe results among the model strategies, while fractional models show lower volatility.
• The comparison omits trading costs and a risk-management process, limiting conclusions about live performance.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.