So, I downloaded the monthly file.[1] I created a rolling 12 month average from 1986 in a spreadsheet and saved to a csv file.[2] I have a general csv function that I use to funnel to two probability R scripts: 1) simple historical and 2) simple Monte Carlo, which just uses 10,000 sims against a mean/standard deviation random selection function based on the historical data.[3-5] I then funneled each period through both scripts, using a bit of trial and error to find the 10%, 50%, and 90% thresholds to generate the following:
Clearly, the simple Monte is a little more conservative in its ranges, so I'll prefer that for the 10-90%. I'll go with the historical mean for the 50%.
Doing a sanity check against the crowd score, I find my numbers to be very different, and I think that's a due to the fact that I'm doing a rolling 12 months from 1986 and not using the numbers provided in the background information. Historically, the recent numbers are high, as you can see from the data in [2].
So, I downloaded the monthly file.[1] I created a rolling 12 month average from 1986 in a spreadsheet and saved to a csv file.[2] I have a general csv function that I use to funnel to two probability R scripts: 1) simple historical and 2) simple Monte Carlo, which just uses 10,000 sims against a mean/standard deviation random selection function based on the historical data.[3-5] I then funneled each period through both scripts, using a bit of trial and error to find the 10%, 50%, and 90% thresholds to generate the following:
date, hist10, hist50, hist90, monte10,monte50, monte90
2022-01-01, 89, 108, 140, 80, 109, 149
2022-07-01, 78, 105, 171, 67, 110, 182
2023-01-01, 69, 103, 197, 60, 110, 204
2023-07-01, 68, 102, 214, 57, 114, 225
2024-01-01, 68, 105, 214, 55, 115, 235
2024-07-01, 66, 104, 214, 57, 115, 241
Clearly, the simple Monte is a little more conservative in its ranges, so I'll prefer that for the 10-90%. I'll go with the historical mean for the 50%.
Doing a sanity check against the crowd score, I find my numbers to be very different, and I think that's a due to the fact that I'm doing a rolling 12 months from 1986 and not using the numbers provided in the background information. Historically, the recent numbers are high, as you can see from the data in [2].
[1] https://www.matteoiacoviello.com/gpr_files/gpr_web_latest.xlsx
[2] https://gitlab.com/cafebedouin/gjp/-/blob/master/data/gpr.csv
[3] https://gitlab.com/cafebedouin/gjp/-/blob/master/csv-gpr.R
[4] https://gitlab.com/cafebedouin/gjp/-/blob/master/functions/simple_probability.R
[5] https://gitlab.com/cafebedouin/gjp/-/blob/master/functions/monte.R
Note to self: I need to rewrite these scripts to automate finding the 50% mid-point and the 10-90% bands because this was a tad tedious.