REPLYOPSAI RESEARCH NOTE
FREQAI · REAL EXPERIMENT

Is FreqAI Profitable?

FreqAI is an ML framework, not a profitable strategy by itself. Profitability depends on the data, features, labels, model, entry and exit logic, costs, market regime and execution.

CURRENT PAPER SAMPLE
+$10.42realized PnL5closed trades2open trades100.0%win rate*
*This is an early PAPER sample, not evidence that FreqAI is generally profitable.
CURRENT FORWARD EVIDENCE

What the experiment actually shows

START $05 CLOSED TRADES+$10.42
EXPECTANCY / CLOSED TRADE+$2.08PROFIT FACTOR∞*MAX CLOSED-PNL DRAWDOWN$0.00

These values use closed PAPER trades only. Open-position PnL is excluded. *A profit factor with no losing closed trades is undefined in practical terms; the displayed infinity marker is not evidence of a stable edge.

Recent closed trades

MARKETSIDEEXITPAPER PNL
SOL09-24 17:40SHORTexit signal+$0.35
ETH09-24 21:32LONGroi+$2.00
ETH09-24 14:55SHORTexit signal+$2.28
SOL09-24 14:50LONGtrailing stop loss+$2.07
ETH09-24 14:32LONGexit signal+$3.72

What FreqAI actually provides

FreqAI adds a machine-learning workflow around Freqtrade: feature generation, model training, prediction and periodic retraining. It does not remove the need for a coherent target, risk controls or realistic evaluation.

Prediction quality and trading profitability are different

Model layer

A model can predict a target better than chance while the resulting trades remain too small, too late or too expensive.

Trading layer

Entries, exits, sizing, fees and simultaneous exposure determine whether predictions become positive net expectancy.

A useful evaluation therefore starts after the model: closed-trade expectancy, payoff distribution, drawdown, turnover and stability across new data.

Why a profitable backtest is insufficient

Model selection and repeated strategy changes can overfit historical data. PAPER trading adds forward observation and operational realism, although it still cannot reproduce every aspect of live fills, latency and slippage.

Our test

V6 currently evaluates BTC, ETH and SOL on a 5-minute timeframe, with a 30-day training window and retraining every six hours. We preserve weak and losing runs instead of publishing only favorable periods.

Open raw experiment summary →

So, is it profitable?

The defensible answer is strategy-specific: FreqAI can produce a trading system, but profitability has to be demonstrated by the individual implementation over a sufficiently large, cost-aware sample. Our current sample is too small for that conclusion.