GM, girls and boys

I’ve been exploring a long-only alpha-generating strategy comparing Triple Exponential Moving Average (TEMA) and Time Series Forecast (TSF) using 6 years of BTC data. So far, it has shown +750% outperformance over simple BTC holding.

Disclaimer: This post is not financial advice. It is shared purely for academic discussion and quantitative research purposes.

Indicators:

TSF (Time Series Forecast)
A moving linear regression using least-squares fit per bar.
Similar in smoothness to moving averages but includes trend.
Good for forecasting and momentum-style entries.

TEMA (Triple Exponential Moving Average)
Attempts to reduce lag by using multiple EMAs: TEMA = 3×EMA - 3×EMA(EMA) + EMA(EMA(EMA))
More reactive to price changes than TSF or traditional EMAs.

This is exactly the kind of indicator logic I’m testing with CorrAI, a no-code quantitative strategy builder that helps backtest and optimize trading strategies using Binance spot data, including BTC/USDT. With CorrAI, I can iterate on TSF vs. TEMA logic in seconds without writing a single line of code.

Strategy Logic (1h timeframe, BTC/USDT)

Return Distribution

Using KDE, the mode of returns is slightly below 0%, meaning the core distribution is mildly negative.
Profitability depends on fat-tail events—any dampening there reduces alpha.
Extreme-loss events are too rare to evaluate cleanly (sample size limitation).

Risk Metrics

UPI: 4.51
VaR: -7.36%
CVaR: -8.73%
Entropic Risk: 2.63
Rachev Ratio: 3.18

CVaR is close to VaR, indicating few heavy-loss events—I didn’t proceed with Conditional Fail Expectation (CFE) or Conditional CFE. The strong Rachev Ratio is a positive, especially since this version doesn’t yet use any dynamic risk adjustment or side constraints.

While the Rachev Ratio looks promising (even without dynamic risk filters), the MDD over 40% is a concern.
I’m seeking improvements that could bring MDD under 10% so that the strategy becomes safer for leverage. CorrAI could help further experiment with stop-loss variants or exposure timing to address this.

Would love to hear your thoughts on:

  • Any ways to improve fat-tail capture or optimize the hold/exposure timing

  • Whether this logic could be extended to LTF (low-timeframe) or multi-asset strategies

Also, if you're into backtesting BTC strategies or building algo-trading logic without writing code, you might want to check out CorrAI. It’s made for exactly this kind of exploration—combining statistical intuition, quant tools, and performance analytics, all through a visual drag-and-drop interface.

Also, I'd love to have your support in my participation in the CorrAI's focus on your trading inspiration event.

Cheers!