No. of Recommendations: 4
overlap list then phl for more to choose 5 stocks every month
Now that is an interesting twist.
Looking at the overlap counts I have been concerned about the number of months where the number of overlaps is just 0 or 1. The count also swings wildly from month to month.
But maybe the times of low overlap are the times when timing has you out of the market, so it's a non-issue?
I ran one recent check where the overlap count was 4. The next check 2 weeks later the count was 2. I would be scared going from, say, 5 stocks at $10K each to 1 stock at $50K.
But maybe the (slight) higher returns of Overlap come from these times you put it all on red? Dunno.
This is effectively doubling up on the PHL-top5 stocks that are also in the RS-top5. You don't need to run a seperate portfolio for overlap, you could just do the overweighting in the PHL portfolio.
Hundreds of hours of conversations with it by now.
Be wary of the possibility of curve-fitting.
"Curve-fitting bias refers to the systemic error that occurs when a mathematical model or trading strategy is excessively tuned to historical data. Instead of capturing the true, underlying trend, the model memorizes specific past data points and noise, losing its ability to generalize or perform well on new, unseen data."
I don't have Norgate data nor have I used Chapgpt, but the biggest value I see here is the evidently confirmation that the GTR1 backtest is valid.
(thus I use 325 days for the SMA of SPY because 300-350 were all good and better than 200 or 253 days), to see if works out-of-sample
Better in what regard? The risk metrics or the CAGR? Better be risk metrics, because that's the benfit of timing.
(BTW, Sortino Ratio is a much better measure than Sharpe Ratio. Sharpe penalizes upside volatility as well as downside volatility.)
Being a simple man, for timing I just use the Yahoo historical price data in my own spreadsheet to get the timing signals and then apply them to the GTR1 daily prices.
We have 40.8 years of prices and 21 overlapped cycles of prices. I have only run the timing backtest on 2 periods, 1985-2026 and 2006-2026. There was no set of timing criteria that worked best for both periods. A better test would be to run the timing backtest on 2 distinct non-overlapping periods, 1985-2006 and 2006-2026.
But.....I don't think it makes any real-world difference.
I tested several lookback periods, various percentages below the SMA for the sell signal, waiting for 2 or 3 or 4 successive weeks of being below the SMA sell cutoff before taking the sell signal, with and without the FRED indexes used by GTT.
Nothing stood out. And none of them were outstanding. What they all did was improve the maximum drawdown, standard deviation, and Sortino Ratio.
Even then, they all failed to reduce the maximum drawdown to a tolerable level. Untimed MaxDD(12 month) was -55%. Timings had -35% to -39%.
That's because of the high volatility. The 3 worst months were -31%, -31% and, -29%. No timing scheme is going to get you out fast enough to avoid those. Of the 490 months, 48 of them had a loss of more than -10%. Timing cut that down to 34.
Lay out the IN/OUT signals by date for a few of the timing schemes. What you see is not major differences. What you see is a one or two month difference in the start of an IN or OUT period. Every OUT period of several months were the same in all the timings, the only difference was slight difference on the start of the period.
Short OUT periods don't do much for you. The value of timing is being OUT during most of the long bear markets.
And, no, checking the timing weekly instead of monthly did not improve things. The week-to-week volatility is also high.
Of the 2128 week, 39 had loss of more than -10%. Weekly timing just had you missing out on the extraordinary high return weeks.
