Comments on: Build Better Strategies! Part 2: Model-Based Systems

By: jcl

jcl — Fri, 16 Aug 2024 09:21:43 +0000

Because we want to run ahead of the phase. We want to catch the cycle before it actually happens.

By: Tom

Tom — Wed, 14 Aug 2024 20:48:03 +0000

For the ‘Cycles’ model, could you explain why you add PI/4 to phase of the sine function? Say if Prices follows exactly that of a Sine function, would you keep the phase unchanged, since the sine function is already model the Prices correctly. Thanks.

By: jcl

jcl — Mon, 22 Apr 2024 08:43:12 +0000

There is no limit to the currency pairs, and backtesting needs no special setup. But you need historical price data of all pairs in your list.

By: Richard Kuama

Richard Kuama — Thu, 18 Apr 2024 09:49:10 +0000

jcl,

I know this is a dated post but – with the price shocks strategy and currency strength functions:
– how many currency pairs can be used/processed?
– Is there any special setup to backtesting using the currency strength functions (separate assets file? all currency pair data present?)

By: Kiran Rama

Kiran Rama — Tue, 04 Jul 2023 14:34:57 +0000

MeanReversion feature including days since peak and valley show very low predictability in ML models. The highest from this was 0.523
very disappointed

Same with ExploitCycles that had a highest of 0.516
in real world on Cross-validated data, these have lower predictability than FMA or SMA also

Ehler’s methods seem overhyped with heavy overfit. No wonder he is doing more workshops than trading

By: rki04013

rki04013 — Tue, 11 Apr 2023 08:41:20 +0000

For FisherN, Zorro uses

// normalize a value to the -1..+1 range
var Normalize(var* Data,int Period)
{
Period = Max(2,Period);
var vMin = MinVal(Data,Period);
var vMax = MaxVal(Data,Period);
if(vMax>vMin)
return 2.*(*Data-vMin)/(vMax-vMin) – 1.;
else return 0.;
}

Why do we normalize as 2*(x-xmin)/(xmax-xmin)-1 instead of just (x-xmean)/sigma

By: jcl

jcl — Wed, 27 Feb 2019 12:45:34 +0000

The threshold is normally not very dependent on volatility since the Fisher transform transfers data to the +-1 range. But it could make sense to slightly optimize the threshold between 0.9 to 1.5 or so, for adjusting the number of trades that the system opens.

By: John Zay

John Zay — Tue, 26 Feb 2019 21:58:50 +0000

Would it be beneficial to fin the p-value of the fisher transform with the Pearson correlation coefficient or am I not approaching the problem correctly?

By: John Zay

John Zay — Tue, 26 Feb 2019 16:58:54 +0000

I have been experimenting with the mean reversion strategy posted on this page. I have been having trouble finding the right threshold for the fisher transform as the numbers increase/decrease based on the volatility of my asset. I was wondering what ways there were to mitigate this issue when using the fisher transform.

By: jcl

jcl — Mon, 28 Jan 2019 03:10:06 +0000

I haven’t written the Z9 system, but know how it works. Main difference to Antonacci is that it trades a portfolio. Assets are ranked and sorted by their dual momentum and the highest ranking assets are traded. Position sizes are determined by relative momentums. There is also a filtering mechanism in place that increases treasure positions when the market tanks.

I believe someone on the Zorro user forum has replicated Z9 to some extent, you could check out his code.