Perpetual futures are central to crypto market structure. Unlike traditional futures, perpetual contracts do not expire. Funding payments help keep the contract price close to the spot market. When funding is positive, long positions usually pay shorts. When funding is negative, shorts usually pay longs.
Funding rates can become useful AI features because they contain information about leverage pressure and positioning. Extremely positive funding may suggest crowded long exposure. Extremely negative funding may suggest crowded short exposure or stress. But the interpretation is not mechanical.
The first research challenge is timing. Funding is calculated and paid on exchange-specific schedules. A model should only use funding information that would have been known at the decision time. If a backtest accidentally uses future funding values, the result is contaminated.
The second challenge is normalization. Funding levels vary by exchange, asset, volatility regime, and market cycle. A raw funding value may not mean the same thing for Bitcoin as it does for a smaller altcoin. Rolling z-scores, cross-sectional ranks, or regime-adjusted measures can be more informative than raw numbers.
The third challenge is interaction. Funding may be more useful when combined with price momentum, open interest, liquidation clusters, volatility, or order book imbalance. A machine learning model can explore these interactions, but the researcher still needs to inspect whether the relationships are stable.
Funding can also be a cost, not just a signal. A strategy that holds perpetual futures must account for funding payments in performance. Ignoring funding costs can make a leveraged strategy look better than it is.
Funding-rate research is best framed as a study of market pressure. It can support hypotheses, but it should not be treated as a standalone trading command.