|
Post by Mustang on Jan 18, 2024 16:00:05 GMT
From what I have read understanding the results of Monte Carlo simulations is difficult. How the results are communicated by a financial advisor can change the outcome of an investor's decision. It's the difference between looking at a cup half full, or looking at it as half empty. One of the articles I read compared the simulations to a weather forecast of a 90% chance of rain. Just because rain failed to fall doesn't make the forecast wrong. It just means that that particular real scenario fell in the 10% of simulations where rain didn't fall. Communication is difficult. It is especially difficult when communicating complicated subjects. In the Wellington/Wellesley Monte Carlo simulation 10,000 scenarios were randomly created. Even though I tried to include a real world SOR some of the scenarios would have left out the 2008 or 2020 market declines (assuming that something like the 2000-2003 downturn was put in the first three years which is unlikely). Softening or leaving out downturns is why 25% of the outcomes had higher ending balances than the real world. Some of the scenarios would have included only the worst downturns repeated over and over again which is why 10% of the outcomes ran out of money. The real world doesn't work that way so both the overly positive outcomes and overly negative outcomes are unrealistic. A real world outcome will be somewhere in the middle which might be why many experts say a 70% or 80% probability of success is acceptable.
Perhaps trying to limit the focus on the negative is why Morningstar used a 90% probability of success. The skeptic in me thinks it was an effort to disprove the 4% rule and get eye catching headlines. But, changing the recommended maximum safe withdrawal rate three times in three years, which might be technically correct, really looked bad.
How do financial advisors who barely understand this themselves communicate it to an investor about to retire? Some aren't. One of the article's said some financial advisors are using fuzzy language when talking to clients to avoid responsibility should the real world fall in that 10%. So what are the recommendations? Most say to use a flexible spending plan. Which basically means Monte Carlo simulations can't determine a safe withdrawal rate. One article specifically mentioned a system like RMDs. Variable of dynamic withdrawal plans are not a bad plan if the investor is cutting back vacations or new cars. They are terrible if the investor needs the money to pay the bills. The authors are saying that Monte Carlo results need to be communicated more accurately. Kitces' once wrote that a 50% probability of success was acceptable if communicated properly. It doesn't mean a 50% probability of failure. It means a 50% probability that changes will need to be made along the way. In my opinion, saying that the withdrawal plan has survived the worst economic catastrophes in history would be easy to communicate. Adding that the future may be different and changes may be needed clarifies what might happen in the future.
P.S. When I say I'm learning something every day I mean it. I think I need to read a little more on Monte Carlo simulations. It's not something I would really use but others use it so it would be nice to understand it a little better.
|
|
|
Post by archer on Jan 18, 2024 23:02:03 GMT
Mustang, I noticed on your MC simulation parameters, the simulation isn't an accurate comparison to real life. By using the starting point of year 2000 and adding the 3 worst years first makes a different scenario than what the sequence turned out to be. 2000-2003 were not the 3 worst years for the PF. Max drawdown years were 2007-2009. Running the MC simulation from 2000 and not entering a return sequence, the 50% came out to $24,934 in 24 yrs.
|
|
|
Post by liftlock on Jan 19, 2024 0:21:57 GMT
|
|
|
Post by Mustang on Jan 19, 2024 4:03:50 GMT
Mustang , I noticed on your MC simulation parameters, the simulation isn't an accurate comparison to real life. By using the starting point of year 2000 and adding the 3 worst years first makes a different scenario than what the sequence turned out to be. 2000-2003 were not the 3 worst years for the PF. Max drawdown years were 2007-2009. Running the MC simulation from 2000 and not entering a return sequence, the 50% came out to $24,934 in 24 yrs. You are correct. Using the three worst years are not the same but I didn't see any other way to add SOR similar to the real world. As far as I could tell it was the three worst years or nothing.
Making no adjustment isn't a comparison to the real world. The real world actually had poor returns during the first three years which random returns wouldn't duplicate. I think the similarity of Monte Carlo's $24,493 after 25 years and Portfolio Visualizer's backtest of $24,482 after 24 years is most likely a coincidence.
|
|
|
Post by archer on Jan 19, 2024 5:10:40 GMT
Mustang , I noticed on your MC simulation parameters, the simulation isn't an accurate comparison to real life. By using the starting point of year 2000 and adding the 3 worst years first makes a different scenario than what the sequence turned out to be. 2000-2003 were not the 3 worst years for the PF. Max drawdown years were 2007-2009. Running the MC simulation from 2000 and not entering a return sequence, the 50% came out to $24,934 in 24 yrs. You are correct. Using the three worst years are not the same but I didn't see any other way to add SOR similar to the real world. As far as I could tell it was the three worst years or nothing.
Making no adjustment isn't a comparison to the real world. The real world actually had poor returns during the first three years which random returns wouldn't duplicate. I think the similarity of Monte Carlo's $24,493 after 25 years and Portfolio Visualizer's backtest of $24,482 after 24 years is most likely a coincidence.
Thanks. I see now my thinking was confused. You are correct that while starting the simulation at Y2K makes for one of the worst 24 year periods largely due to 2000-2003, the simulations are still random within that time period and not putting those years first.
|
|
|
Post by Mustang on Jan 19, 2024 17:46:47 GMT
Interesting article. He doesn't seem to like Monte Carlo simulations very well. Below is a summary.
Monte Carlo simulations add random fluxuations to steady growth. Its a step forward from standard steady growth calculators but has serious flaws. MC simulations are based on statistical randomness along a straight line that is generated using a distribution curve. There are many distributions curves. The question is which ones best fit retirement plans. In real life distribution curves are different and change shape over time.
The entire idea of randomness is a flaw. Random outcomes ignore secular trends. Markets may be random short-term but tend to trend long-term. There are three separate trends: bullish, sideways, and bearish each with a different distribution curve working at different times. Trends can last up to 20 years. Randomness ignores the correlation between market events. The sequence of market events are not random. Good market results drive markets higher. Bad ones can cause waterfalls or flat markets. Monte Carlo simulations rarely have multi-year bull or bear streaks
It appear that the simulations have little to no relationship to the real world. Otar then listed recommendations for a better simulator.
|
|
|
Post by Mustang on Jan 19, 2024 21:05:17 GMT
I went back and read a couple of articles. It is clear that Monte Carlo simulations were meant to be used by financial advisors, not investors. This is a summary of Robert Powell's article. www.marketwatch.com/story/should-you-use-a-monte-carlo-simulation-to-determine-if-your-retirement-savings-will-last-11611607222Monte Carlo simulations were thought to be a better approach because the computer was running thousands upon thousands of scenarios. He said advisors need to stop talking Monte Carlo results. Basically, people don't understand it. Monte Carlo simulations are best used to test portfolio allocation recommendation not for client presentations. It can provide a view useful for back of the envelope planning but the average path is of little use when the client only gets one shot at retirement. Worst paths can be used for planning - less for what may happen and more for planning how to respond to the arrival of bad outcomes. Average path is of little use. If your plan can succeed in the worst case scenario that should provide some degree of comfort. (Comment: Didn't Bengen say that in 1994 and its now repeated in 2021.) He said there is good news. Just like Bengen's 4% Rule was replaced (?) Monte Carlo may be replaced. The tool is VAR (value at risk). It's outcomes are in dollars not percents and probabilities. Here is a summary of Massima Young and Wade Pfau's article: www.advisorperspectives.com/articles/2023/01/10/the-dangers-of-monte-carlo-simulationsResults are heavily dependent on capital market assumptions. Probability of success is the percentage of simulation successes. Each year of the simulation is assigned a return that is randomly selected from a distribution curve determined by assumed returns, volatility, and correlations between asset classes. The wide variation of these assumptions was mentioned in a previous post. What can advisors do? Use the average of different capital market assumptions. "Wisdom of the crowd." Stress test the portfolio using different capital market assumptions by lowering predicted returns, increasing volitility, and using more positive correlations. But while a stress stress test give greater confidence it also leads to under spending. Other advice: Rely on other options such as income flooring, bond ladders and annuities to secure cash flow. Other options that do not rely on future forecasts. Mustang comment: It seems to promise far more than it can deliver. I'll stick with historical data.
|
|
|
Post by archer on Jan 20, 2024 1:13:43 GMT
Young and Pfau's article is seems to be in the context of advisors using MCs to show their clients a safe outcome. For people who are turning their PFs over to their advisors and don't want to be proactive, basically hoping their advisors have a crystal ball. The problem isn't so much Monte Carlo, but rather what is expected of it.
|
|
|
Post by bb2 on Jan 20, 2024 1:49:17 GMT
Like I said. Sales tool. Beware the financial industrial complex.
|
|
|
Post by Norbert on Jan 22, 2024 17:45:04 GMT
Thinking about using Monte Carlo simulations to help with portfolio design and withdrawal rates, I suggest setting the Sequence of Returns to "Worst 5 Years first" to really stress test your plan. This might reflect, to a degree, what might happen if a real war breaks out. What are the odds of that? Low, probably, under 5%. But, it's a tail risk that would kill the markets for a while, while also offering new opportunities (if we survive the conflict). My strategy is (1) keep fixed expenses low, so I can stop all withdrawals for a while by just cutting back on discretionary expenses; (2) keep a significant part of the portfolio in safe assets. FWIW, N.
|
|
|
Post by Mustang on Jan 22, 2024 18:36:14 GMT
Thinking about using Monte Carlo simulations to help with portfolio design and withdrawal rates, I suggest setting the Sequence of Returns to "Worst 5 Years first" to really stress test your plan. OK. 69% probability of success. After 30 years $10,000 dropped to $7,600 at the 50th percentile. Half the simulations better. Half worse. 25% ran out of money before 22.5 years. 10% of the simulations ran out of money before 17 years. At the 50th percentile the safe withdrawal rate is 4.58%. I can make that work. All it takes is a little tweaking. Use a 4% initial withdrawal and take 1% point less than inflation increases. For example, if CPI is 3% only increase the annual withdrawal 2%. www.portfoliovisualizer.com/monte-carlo-simulation#analysisResults
Since half the simulations are better it might be possible to increase withdrawals later. Kitces' once wrote that if the ending balance of the portfolio was 150% its beginning balance it is safe to increase the withdrawal 10% but that should be done only once every three years. That is one method Since studies have shown 100% historical success for 4%-30 years, 5%-20 years, and 6%-15 years, I'd calculate mid-points. For example, with 27 years to go it would be around 4.33%. For example, the the withdrawal was still 4% at the 20 year point I'd give myself a raise and then keep the annual inflation adjustment at one point less than CPI.
A more realistic payout period would be 20 years. Planning for 30 gives a little extra padding.
|
|
|
Post by archer on Jan 22, 2024 18:46:29 GMT
Interesting thought Norbert , It is hard to predict the effects of war on the market. I'm sure it would depend on the war and the outcome. I looked back at the market during WW2 and see, 1939 -5.45% 1940 -15.29% 1941 -17.86% 1942 +12.43% 1943 +19.45% 1944 +13.8% 1945 +30.72% WW2 started in 39 and the US had 3 bad years. We entered the war in Europe in Dec '41 and apparently investors were very optimistic about our victory. '46 and '47 were bad years, with low employment, and less support for transitioning back to peacetime manufacturing. If major war on the scale of WW2 were to happen again, I don't think the US economy would shift to war economy in the same way. We had a predominantly manufacturing economy which transitioned to a war economy, tanks, ships, weapons etc. Now we have much more of a service and tech economy. More importantly, what manufacturing we have left, is now being done by countries we might be at war with, or at least not be supportive. 5 worst years for sequence of returns is probably about right, but might be out of the range of possibility for many of us.
|
|
|
Post by bizman on Jan 22, 2024 21:24:44 GMT
Interesting thought Norbert , It is hard to predict the effects of war on the market. I'm sure it would depend on the war and the outcome. I looked back at the market during WW2 and see, 1939 -5.45% 1940 -15.29% 1941 -17.86% 1942 +12.43% 1943 +19.45% 1944 +13.8% 1945 +30.72% WW2 started in 39 and the US had 3 bad years. We entered the war in Europe in Dec '41 and apparently investors were very optimistic about our victory. '46 and '47 were bad years, with low employment, and less support for transitioning back to peacetime manufacturing. If major war on the scale of WW2 were to happen again, I don't think the US economy would shift to war economy in the same way. We had a predominantly manufacturing economy which transitioned to a war economy, tanks, ships, weapons etc. Now we have much more of a service and tech economy. More importantly, what manufacturing we have left, is now being done by countries we might be at war with, or at least not be supportive. 5 worst years for sequence of returns is probably about right, but might be out of the range of possibility for many of us. Don't forget that we won WW II without any damage to our cities and manufacturing base. A potential lost major war to China and friends, without the benefit of a Pax America and Marshall Plan with liberal trade help to magnanimously rebuild the losers as Japan and Germany got in the aftermath of WW II could be breathtakingly bad for us. That's why I hope we do enough to deter Xi. It will cost us, but not remotely as much as losing a major war.
|
|
|
Post by richardsok on Jan 23, 2024 15:53:56 GMT
I'm going to assume that dollar erosion going forward is a certainty. With that in mind I get good results backtesting a simple portfolio of 96% ITOT and 4% NUGT, even allowing for inflation and 5% withdrawals. Has anyone put it thru a MC simulation?
|
|