When doing a historical analysis, the output I'm seeing in the main graph on the Run Analysis page does not seem intuitive to me, vs how I understand the MC situation to work.
I've entered in all my income/spending. Then on the run analysis page I select specified expenses with smoothing. I select historical for the solution type, then perform the smoothing. Then I go up and select historical results (make it red) and then click Update Active Analysis. That screenshot is attached.
First off, I would have thought the red line would be the median. But it's not? It seems it's still displaying the MC median.
Next, I though consumption smoothing with the historical setting would generate the 10th percentile case, similar to MC. But in this case there are no failures. When I show 1966 as a historical start year, it indeed doesn't fail (but almost). So my question is, what do the percentile bands represent? Relatedly, how is the smoothed spending amount determined?
With MC based CSS, the 10th percentile band is always close to the x-axis, as I would expect.
Finally, when I compare my historical analysis with my MC, there is a huge difference (MC shows 4-fold lower terminal values). Yes, I'm using 4% real return (with 3% inflation) for stocks and 1.% for bonds in my assumptions for a stock heavy portfolio. So perhaps all that means is historically the markets did much better than my assumptions. However, my smoothed spending for the MC and historical runs are nearly identical (within 2%). This doesn't make sense to me. If my MC assumptions are so much worse than history, why is the smoothed spending the same?
@stkeros The red line on the graph on the Run Analysis page always represents the deterministic projection based on your specific rate of return (ROR) inputs; it never represents the median of either Monte Carlo or historical analysis results. The randomized RORs of a Monte Carlo analysis are based on your ROR inputs, so there's good correlation between the red line and the blue bands of the MC analysis; however, historical analyses totally ignore your specific ROR inputs and use the historical RORs for your asset classes and, therefore, can yield blue bands that are much more optimistic as in your screenshot.
Consumption smoothing done with the historical analysis method runs repeated historical analyses until it finds a solution with roughly a 90% success rate; however, it does have a maximum number of iterations it will run before it stops, so it may not be finding a 90% solution before stopping. I agree that it’s very suspicious that the MC and historical methods would return very similar consumption smoothing results, so there may be something else going on. I’d be happy to take a look at this and try to provide a better explanation if you’d like to share an export file (if so, please send to mail@pralanaconsulting.com).
Thanks,
Stuart
Thanks stuart, I get now the red line is always deterministic (which I knew and forgot). But I had meant to include a screenshot which also had a yellow single-year line. E.g. the years 1961 through 1970 or so all fail well before the end of the plan. Yet the lower edge of the blue band is well above the x-axis when including the historical smooth spending as an essential expense. Maybe the few bad years were really bad, but less than 10% of all the years failed and thus the 90% band doesn't approach the x-axis? But then again, I would have assumed the historical smoothing would have been set such that it would "pull" the results down (similar to the way MC works where the 90% band is near the x-axis at the end of the plan). I'm sending you the data file by the way.
Thanks for sharing your export file. I confirmed that the historical smoothing algorithm is indeed stopping due to reaching the iteration limit rather than achieving a 90% success rate; however, on the final iteration, the annual spending was about $259K and the adjustment amount was only $1000, and this yielded a 94% success rate. So, it is very close to converging on a 90% solution. The Monte Carlo analysis yielded a 94% success rate with an annual spending amount of about $236K. The historical spending amount is just $23K higher than the MC amount, but this is spread over a 50-year planning horizon.
The summary question is why the terminal amounts of historical vs. Monte Carlo analyses are so different with such a seemingly small difference in annual spending: the 10th-90th percentile range for historical analysis is roughly $4M-$30M while the 10th-90th percentile range for MC analysis is $1M-$12M. Given the 50-year planning horizon, this plan is very sensitive to annual spending: using deterministic analysis (which uses the RORs and you specified on the Financial Asset page), final savings is about $330K with annual CS spending of $259K whereas final savings is about $3.7M (a factor of over 10X higher) with annual CS spending of $236K. You specified a nominal ROR for stocks of 7% (which is the mean for the MC analysis) while the historical analysis used the historical mean of about 11.5% for stocks, so the historical smoothing analysis yields about $23K more in annual smoothed spending. That’s roughly 10% more spending over a 50-year period based on the more-optimistic RORs for historical analysis. Another interesting observation is that when I do an MC analysis using the $259K of annual smoothed spending (yielded by historical CS), the plan’s success rate dropped to 70%. When I do a historical analysis using the $236K of annual smoothed spending (yielded by MC CS), the plan’s success rate goes to 100%. So, in conclusion, the terminal amounts are very different due to the length of the planning horizon and the significant difference in market optimism between the two analysis methods.
Stuart
