I spent a few hours working with ChatGPT Plus to assemble historical data for the following asset classes:
Long-term treasuries
Large-cap growth
Small-cap value
Gold
Managed futures (actual data only went back to 1980, so I just had ChatGPT extrapolate the values back to 1928).
It appears I can only attach one file, so I will make 5 posts to attach all the files.
Once again, there is only data for managed futures starting in 1980, so I had ChatGPT extrapolate the existing data back to 1928 ... better than nothing.
Here are some charts from Paul Merriman. 90 years (30 x 3) return of S&P500, US large value, small balanced and small value.
Thanks for sharing that, though I’m not sure how to use it within Pralana. Pralana lets you define custom asset types as long as you can supply annual return histories. That’s why I posted TXT files with year-by-year performance for multiple asset classes—to make historical modeling possible beyond Pralana’s limited built-in datasets.
I’m not a strong fan of Monte Carlo, which assigns random returns to each asset across thousands of runs; assets rarely all rise or fall together, so many simulated paths are unrealistic. By loading actual histories for my user-defined assets, the historical backtest feels more credible. I can also start the test in a known “worst” retirement year to stress sequence-of-returns risk and specific economic regimes. Bottom line: the historical approach is more robust and gives me greater confidence after seeing how the portfolio behaves across real-world conditions.