Skip to content
All library documents

Testing and Combining Factors in Crypto Perpetual Futures

Article FMZ digest · Author: 发明者量化-小小梦

Summary

This beginner-oriented article outlines a cross-sectional factor research workflow using hourly Binance USDT perpetual futures data. It describes collecting price, volume, trade-count, and taker-buy information across many contracts, then forming candidate factors to rank currencies. Factor validity can be assessed through forward-return regression, information coefficient and information ratio, or grouped portfolio backtests that test whether returns vary consistently across ranked groups. Before combining factors, the article recommends clipping extreme values, standardizing each factor, and handling missing data. It then illustrates a weighted composite and mentions historical-return weights, IC-based optimization, and principal component analysis as possible combination approaches.

The text is a framework and partial demonstration, not a complete empirical study. Some factor definitions and results are omitted, and the supplied composite weights are illustrative without enough supporting statistics to judge performance. The data sample and market universe are limited to the described exchange and period, while survivorship, transaction costs, liquidity, and out-of-sample validation are not adequately addressed. Its central lesson is the need to continually re-test factors because their effectiveness can decay as markets and capital flows change.

Key ideas

  • The proposed research universe consists of multiple crypto perpetual contracts and uses price and trading activity data.
  • Forward-return regression, IC and IR measures, and ranked-group tests are offered as ways to evaluate factors.
  • Factors should be standardized and cleaned before they are combined because their scales and distributions differ.
  • Composite factor weights can be based on historical returns or IC behavior, while PCA offers a dimensionality-reduction alternative.
  • Factor effectiveness can decay, so findings require ongoing validation and the example does not establish robust live returns.

Tags

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