I've seen the thread where historical and monte carlo results were discussed, and several people seemed to say that they got better results from Monte Carlo than from Historical. I get better results consistently from historical, and want to understand if I am setting it up correctly. I am not using CAPE, I have a start year of 1928, and and I have not checked the box for "Activate Historical Sequence Analysis" because I don't understand it, and haven't had time to see if it merely adds information or would change the result. When I set up Monte Carlo, I have checked the box for Correlated ROR's but not Custom std deviations. I seem to remember a setting somewhere for something called "deterministic" but can't remember where or what is was.
Can someone help me with the set-up parameters for these things? My wish on historical would be to have the broadest set of data possible, which is why I picked 1928. I want the most conservative results, so that hopefully I wouldn't have unpleasant surprises. Pralana is a very powerful tool, and I don't want to cause GIGO errors by not understanding how to use these parameters, even though I don't really understand what they all mean.
So sorry, this is a duplicate post.
Not sure what you mean by getting "better" results with one vs the other of Historical vs. Monte Carlo. They are different ways of making guesses about the future. Personally, I use historical.
Pralana only uses data back to 1928, which they get from published data from NYU's Professor Damodaran. Researchers like Shiller have gone back to 1871, but you have to realize that the data quality gets worse and worse as you go back in time and you have to wonder if the era of Civil War Reconstruction, recent completion of the Transcontinental Railroad, no electricity and few regulations really have much to do with future markets anyway. Even bond yields are largely reconstructions prior to the 1950's as there weren't regular auctions of 10 year Treasury bonds.
Generally, I tune my deterministic results so that they correspond to the 20-30th percentile of historical results (meaning 70-80% of historical cases were better).
The historical sequence is kind of cool as it lets you see the ups and downs in the market in numbers in your accounts, not just a line on the graph. Also, there are 10% of the historical starting years worse than the lowest line on the graph, so sometimes it's worth looking at how bad it has gotten before - the default bad starting year is 1965 as that was the worst in Pralana's data. I like to use the historical sequence to stress test my Roth Conversion plan. I will disable Roth Conversions, then look at a bad historical starting year, and note whether the plan survived and the final account values. Then I enable Roth Conversions (then you have to return to Historical Analysis and reselect the historical sequence), and see how wha the Roth Conversions did (i.e. in bad historical sequences, Roth conversions make things worse). It's a sobering reminder that the benefits of Roth Conversions are based on projections of growth that is uncertain to occur.
Not sure what you mean by getting "better" results with one vs the other of Historical vs. Monte Carlo. They are different ways of making guesses about the future. Personally, I use historical.
By better I mean that I get results of 100% with historical, and somewhat less with Monte Carlo. This seems be different than what several other forum posters have noted. Further reading has informed me that it's likely because my rates of return on set-up might be lower.
When you say you tune your deterministic results, how do you do that? I saw mention of "deterministic" a while back, looked it up, and then forgot what it meant, so I'll repeat that. But I'm curious how about how to "tune" for them.
"Deterministic" means the rates of return you tell Pralana (either under Build-Simple Portfolio Modeling-Rates of Return or Build-Advanced Portfolio Modeling-Rates of Return). Since those returns are really just your guesses about the future, Historical and Monte Carlo add value by letting you test what might happen in different circumstances like a bad sequence of returns.
I set real bond returns at the current yield of 20 year TIPS (2.36% currently), since you can get that guaranteed, and then iterate on stock returns until the deterministic answer falls on the 20-30% percentile of historical results.
