6 Comments
User's avatar
Jalal's avatar

In your experience, how closely does the actual performance of a strategy as such (pops) track the backtest results when implemented in real-time trading? Does it tend to be closer to the YC/TO approach or the canonical close-to-close approach?

Expand full comment
Jaewon Jung's avatar

It's hard to tell, especially for mean reversion pop strategies given their sporadic nature. You probably need a few years worth of data to get to any statistically meaningful conclusion.

Expand full comment
hydromod's avatar

One strategy I've seen for the overbought 10-day TQQQ RSI swaps in UVXY. The minimum threshold for expected gain seems to be around 79%.

Although there are at least two definitions for calculating RSI that I've seen. The one that strictly uses the last ten days seems less reliable than the one that acts like a type of moving average.

Expand full comment
Jaewon Jung's avatar

Yeah, investing in UVXY when overbought will produce a better return albeit riskier. By the different RSI calculations, do you mean the look-back period used when calculating the EMAs for average gain / loss of an RSI?

Expand full comment
hydromod's avatar

Same nominal 10-day lookback. One uses just the 10 days. The other one updates the previous RSI using ((N-1)/N) of the previous-day totals and adding the latest day. That second one seems to have more reproducible thresholds but is a bit starting-point dependent, as it has a bit of extended memory of all previous days.

Expand full comment
Jaewon Jung's avatar

Yes, that's what I meant by the "look-back period": how long you look back to compute the initial (seed) value for weighted averages. For instance, for 10d EMA, a shortcut is just using 10d SMA (i.e. (Pt-10 + Pt-9 + ... + Pt-1) / 10) as the seed, which is not ideal. QuantMage takes all the available history into consideration to seed it, not only for backtesting, but also for live trading :)

Expand full comment