Discrepancy in Forecast Scalar ?

[ea1442]

Hello,

I just started reading the documentation and trying to get familiar with PST.
In the Introduction, the forecast scalar for EWMAC32 is calculated as follows:

2016-05-05 2.846447
2016-05-06 2.846448
2016-05-09 2.846442
2016-05-10 2.846438
2016-05-11 2.846425
Freq: B, dtype: float64

However, when I run the example code in the same sequence, the forecast scalar for EWMAC32 returns as:

2023-08-25 8.691212
2023-08-28 8.691153
2023-08-29 8.691078
2023-08-30 8.690987
2023-08-31 8.690886

To make sure we are looking at the same dates, I checked with
print(my_system.forecastScaleCap.get_forecast_scalar("SOFR", "ewmac32").loc["2016-05-11"])
The result is : 8.786838429592347

I also ran the example code shared on [GitHub.
The forecast scalar for EWMAC was again around 8.6x.

In the rest of the code, we override the scalar as 2.65, as suggested in Rob's books. It would make sense if the calculated forecast scalar were around 2.846425 as shown in the introduction notebook. However, with the current forecast scalar being around 8.69, wouldn't this adjustment be incorrect?

What am I missing here?

Thank you.

[vishalg]

How many instruments do you have in the backtest? I believe the forecast scalar estimates are pooled across instruments, if SOFR is the only future in your backtest, I am not surprised that your forecast scalar is so high. SOFR vol is very low which will bias the scalar higher.

[ea1442]

Used instruments were : "SOFR", "US10", "CORN", "SP500_micro"
I was just following the steps in the Introduction document.

[robcarver17]

"SOFR vol is very low which will bias the scalar higher." I don't think it will.

[vishalg]

Fair enough. Trend signals are scaled to a vol target so that doesn't make sense. It is probably more to do with the relative short history which means we just don't have enough data to get a robust estimate.