|
Post by yogibearbull on Jan 7, 2024 1:34:50 GMT
Asset Allocation & Withdrawal Strategies in Retirement – Bolin There is an interesting article by Charles Lynn Bolin (@lynnbolin2021) on withdrawals in the January 2024 MFO issue. In the SUPPLEMENTARY information presented here, portfolio # won’t be used as they can cause confusion. In Bolin’s Table #1 with 4 portfolios tested with 6% withdrawal rates, be aware that those are 6% withdrawals from the YEAREND balances every year. So, the amounts withdrawn will fluctuate widely with the market. The residual balances are shown in Figure #1 to indicate whether those kept up with inflation. Interestingly, PV has a parameter PWR (under the Metrics tab) that provides max % withdrawals from YEAREND balances that will also leave inflation-adjusted residual at the end. A reference is made to “the time-tested 4%”, but that so-called Bengen’s Rule has a different withdrawal regime – 4% of the INITIAL lump-sum that is subsequently adjusted annually for inflation; the PV run settings also allow for this and the related PV metric SWR is the max % withdrawal in this scenario (that will EXHAUST the portfolio). One can also deduce from the PV run data the max % of the INITIAL lump-sum that is subsequently adjusted annually for inflation AND leaves inflation-adjusted initial lump-sum at the end, and that % is SWRM. The table below shows PWRs, SWRs, SWRMs. Portfolio; PWR; SWR; SWRM 50%VFINX + 50%VTRIX; 5.59%; 7.71%; 6.79% 70%VFINX + 25%VBMFX; 5.92%; 7.77%; 6.96% 50%VFINX + 25%VTRIX + 25%VBMFX; 5.40%; 7.46%; 6.50% VFINX; 7.06%; 8.77%; 8.19% Bolin’s conclusions about inflation adjusted residual balances are consistent with whether the PWR is less than, or greater than, 6% in the table above. It is also interesting that all could support Bengen-style 4% (w/COLA) and even higher (note SWRs). In fact, all would leave more than inflation-adjusted initial lump-sum even with 6% w/COLA (note SWRMs). Of course, all the data and conclusions are for the period 01/1987-12/2023. www.mutualfundobserver.com/2024/01/asset-allocation-and-withdrawal-strategies-in-retirement/
|
|
|
Post by Mustang on Jan 7, 2024 10:46:49 GMT
I agree with Bolin’s conclusion that a balanced portfolio is needed in the withdrawal phase (decumulation) of investing. While I thought that changing the equity allocation up and down from 35-65% due to market conditions was interesting. I’m not sure how practical it is.
Assuming retirement at age 70 the asset allocation is 65/35. He makes some big swings when he is 75 and again at 83. He makes several smaller swings from 65% equity down to maybe 55-60% (as if that would make a big difference in outcome). Then in his 90s he makes some big swings again. Does he really believe there won’t be any mental decline?
He also says he follows the bucket approach. Looking at the complexity of his charts I’m sure that means multiple assets in each of the three buckets and active management pruning back the winners all the way to the ripe old age of 100.
Sorry. I don’t believe it. His system is too complex for an aging retiree to manage. There will be a point in time where management of our portfolios will have to be passed on. Then our successor will make all the decisions.
To be honest, I’m wondering what our options are when that happens to us.
|
|
|
Post by yogibearbull on Jan 7, 2024 13:35:24 GMT
Bolin writes monthly features for MFO. They are typically long, organized in sections, include lots of general data, and also what he may be doing with his personal portfolio. My comments above related only to Section 3 of his article in January 2024 issue of MFO. They are also related to several posts here on withdrawals, e.g. Horse Race, etc. Mustang has picked up on other aspects also. big-bang-investors.proboards.com/thread/2894/horse-race?page=5&scrollTo=44984
|
|
|
Post by FD1000 on Jan 7, 2024 15:30:08 GMT
I agree with Bolin’s conclusion that a balanced portfolio is needed in the withdrawal phase (decumulation) of investing. While I thought that changing the equity allocation up and down from 35-65% due to market conditions was interesting. I’m not sure how practical it is. Assuming retirement at age 70 the asset allocation is 65/35. He makes some big swings when he is 75 and again at 83. He makes several smaller swings from 65% equity down to maybe 55-60% (as if that would make a big difference in outcome). Then in his 90s he makes some big swings again. Does he really believe there won’t be any mental decline? He also says he follows the bucket approach. Looking at the complexity of his charts I’m sure that means multiple assets in each of the three buckets and active management pruning back the winners all the way to the ripe old age of 100. Sorry. I don’t believe it. His system is too complex for an aging retiree to manage. There will be a point in time where management of our portfolios will have to be passed on. Then our successor will make all the decisions. To be honest, I’m wondering what our options are when that happens to us. Bolin is a pretty good analyst, but the average Joe investor would not be able to come close. I like his big swings, after all, I have done it since 2000. Small swings are mostly negligible and there is no need to complicate stuff with buckets and/or many funds. I found a "scientific" proof how to check someone's mental capacity. All you got to do is watch how someone plays Bridge. A good player needs memory, strategy, analyze ad hoc situations, and boldness. I had the honor to play with/against a very good player for 15 years (see died close to age 100). She was as good as she was at age 85-90, and much better than most.
|
|
|
Post by archer on Jan 7, 2024 16:56:50 GMT
I don't know how to play bridge, so I guess I'm already incapacitated. Chess is another gauge perhaps. Memory isn't involved much, but the ability to mentally navigate multiple future consequences is important. I suck at chess, but as with anything, it takes some study and practice to realize the extent of aptitude, or the point at which reach a plateau of improvement. BTW, my ex mother in law, is now 95 and still playing bridge. At that age, still being able to enjoy it is a good sign of mental health.
|
|
|
Post by FD1000 on Jan 7, 2024 17:39:34 GMT
I don't know how to play bridge, so I guess I'm already incapacitated. Chess is another gauge perhaps. Memory isn't involved much, but the ability to mentally navigate multiple future consequences is important. I suck at chess, but as with anything, it takes some study and practice to realize the extent of aptitude, or the point at which reach a plateau of improvement. BTW, my ex mother in law, is now 95 and still playing bridge. At that age, still being able to enjoy it is a good sign of mental health. I generally never liked card games because most involved luck, Bridge is mostly about skill. I use to play chess too. I join my city chess club at age 13, at age 15 I won it and then I quitted because it wasn't fun for me. Only 2 people play, you need to be quiet and concentrated all the time. Bridge is more friendly, you can talk and socialize, you switch partners and other players every 30 minutes(we play 5 hands per round), you start a new game every 5-7 minutes, you can come out at the top even if you made a few mistakes, the right boldness helps you. I enjoy Bridge very much but I like to win too and I do. Probably 50+% of players just show up to socialize and never improve, they just don't care. Most of the best players I met are number oriented. Bridge opens the door to lots of friends and hours of fun and FOOD(by the players) for just $1 for 3-4 hours of play at 3-4 city adults center around us. We also organize playing at private houses with people we really like. Why not learn it starting next week? The best/easiest starting book is Bridge for dummies, used + shipping under $6 ( link) ( link). Then just start playing, in just 2-3 weeks you can be better than 20-30% of players. You can get better playing live online Bridge 24/7 for free with players around the world and how I got better, see BBO( www.bridgebase.com/).
|
|
|
Post by fritzo489 on Jan 7, 2024 20:43:28 GMT
Mustang, assuming retirement at 70, that's a lot to assume !
|
|
|
Post by fritzo489 on Jan 7, 2024 20:51:02 GMT
FD1000, so FD if you're that good at chess, why not play multiple players (games) at the same time ?
|
|
|
Post by FD1000 on Jan 7, 2024 21:34:33 GMT
FD1000 , so FD if you're that good at chess, why not play multiple players (games) at the same time ? I was starting to be good, I haven't played much for decades, and when I did I was bored. I preferred to play 4 hours of soccer and basketball when I was 15.
|
|
|
Post by archer on Jan 7, 2024 22:23:39 GMT
My daughter was gifted with intelligence, found school and AP classes to be very easy. I taught her to play chess when she was 7. After the 2nd game I never was able to beat her, LOL! I mostly have an oversight problem causing me to make stupid moves. Recently I played online for a few games for about a week and got better at it. Interesting how making careless moves can be outgrown.
|
|
|
Post by FD1000 on Jan 7, 2024 23:04:12 GMT
Chess, Bridge, and others in this category: to be good, you have to study from books/articles/videos, be able to analyze new situations fast, pay attention to what others do, and practice, practice, practice. During 2020 (covid) I played mostly online at BBO. I made sure to follow the real experts and I learned a lot very quickly. This is the beauty of this site, if you don't play you can watch others, it's a great way to get better. I have selected activities that I feel good about and I can do well. Some people are not good with numbers but much better in others. I'm terrible with learning languages, no way I'm learning a 3rd one, especially after English.
|
|
|
Post by Mustang on Jan 8, 2024 2:29:44 GMT
Mustang , assuming retirement at 70, that's a lot to assume ! Full social security is around 67, RMDs were required at 70 are now at 73. And, Brolin's life expectancy chart ran up to 100. I thought it was a fair assumption for our discussion..
But you are correct in a way. I used the word "retirement" when I should have said "payout period begins." Its possible to retire without withdrawing money. If the payout period began at age 65 instead of 70 a 30 year payout period still requires active management into the 90s.
Note: The FIRE (Financial Independence Retire Early) movement is all about early retirement. But, in Bolin's life expectancy chart only one line started at age of 61. The other three were later so I didn't think FIRE applied.
|
|
|
Post by Mustang on Jan 10, 2024 16:30:42 GMT
I read this in another thread. It seemed to apply. "There are a few lessons to draw from the 60/40 portfolio’s swift fall and partial redemption. For one, it’s not realistic to expect any portfolio strategy to excel in every market environment. As my colleague John Rekenthaler points out, it’s easy to criticize traditional balanced funds for not adapting their portfolios to changes in market conditions, but it’s far more difficult to craft something that works better. Market shifts—especially fundamental regime changes—matter, but how to position a portfolio in response is usually only obvious in retrospect." www.morningstar.com/portfolios/why-naysayers-were-wrong-6040-portfolioI would never recommend that low an equity position during the accumulation phase but for a retiree sequence-of-return risks are real and we cannot see the future. I just think that Bolin made something that could be simple into something that was unnecessarily complicated.
|
|
|
Post by FD1000 on Jan 10, 2024 18:03:25 GMT
Mustang="I just think that Bolin made something that could be simple into something that was unnecessarily complicated." FD: the above is true for the accumulation phase too.
|
|
|
Post by archer on Jan 16, 2024 3:08:28 GMT
Monte Carlo simulations are used for assessing probability of PF survival. I know they don't predict the future, but rather run thousands of return scenarios, based on hypothetical possibilities given the characteristics of the assets used for the simulation.
Has any lookback ever been studied to assess how probable the probabilities are? I can imagine this would be kind of complicated, but theoretically a large quantity of old simulations could be compared to the actual outcomes. Eg. if the 10% probabilities occured 10% of the time, that would be a verification of how effective the MC simulations are. THis would be a matter of backtesting many simulations for different tickers.
|
|
|
Post by richardsok on Jan 16, 2024 12:39:17 GMT
Monte Carlo simulations are used for assessing probability of PF survival. I know they don't predict the future, but rather run thousands of return scenarios, based on hypothetical possibilities given the characteristics of the assets used for the simulation. Has any lookback ever been studied to assess how probable the probabilities are? I can imagine this would be kind of complicated, but theoretically a large quantity of old simulations could be compared to the actual outcomes. Eg. if the 10% probabilities occured 10% of the time, that would be a verification of how effective the MC simulations are. THis would be a matter of backtesting many simulations for different tickers. I agree about Monte Carlo studies and backtesting, archie. Traders, like gamblers, do develop cognitive biases such as the gambler's fallacy (belief that past events influence future outcomes) or the illusion one has more control over outcomes than is realistic. Backtesting & MC have their uses but can never allow for black swan events. I believe the posters on this forum are a cut (or two) above the general run of retail investors, but none of us are immune from subtle biases. Further, past investing environments have never included such a colossal debt overhang that is, in fact, accelerating. I have no idea what the consequences may be, but doubt they will benefit anyone's TR.
|
|
|
Post by Mustang on Jan 16, 2024 19:26:04 GMT
There has been some comparisons between history and computer simulations. Most trying to disprove the 4% Rule. Monte Carlo simulations has a lot of advocates but it has some serious limitations. “[Monte Carlo, advocates] argued, was an improvement over Bengen’s approach because a computer was running thousands upon thousands of scenarios on your portfolio. And these scenarios would stress test your portfolio using various market returns, inflation rates, withdrawal rates and the like… It might work when it comes to saving for retirement, less so in the drawdown phase or what some call the decumulation phase.” “Accumulating wealth is a linear process and predictable, but in the nonlinear world of retirement income, returns and standard deviation are not predictors of success and therefore are unreliable inputs for Monte Carlo analysis,” wrote Sandidge… Retirement income, said Sandidge, is governed by chaos theory, which makes it unpredictable.” “Others agree. Generically, Monte Carlo analysis requires laying out the probability structure for the way that the future will evolve, said Michael Zwecher, author of Retirement Portfolios… he said. “What isn’t thought out beforehand is omitted; but like color blindness, the analyst may not be aware of what they’re not seeing.” “Monte Carlo is heavily dependent on the inputs provided, including capital market assumptions, asset class correlations and retiree longevity, said Nersesian. “A small change in any of these inputs can produce a significantly different projection,” he said. “Think ‘garbage in, garbage out.’” “Monte Carlo is just a very laborious sampling exercise, said Nikolic. “And the more you try to capture with a single model, the more you’re susceptible to a model error… and the more you try to accomplish the worse it is.” “In an interview, Sandidge also said he thinks the average path is of little use. He recommends examining your probability of success given a worst-case scenario. And if your plan can succeed in the worst-case scenario that should provide some degree of comfort.” Reference: www.marketwatch.com/story/should-you-use-a-monte-carlo-simulation-to-determine-if-your-retirement-savings-will-last-11611607222“… the results of Monte Carlo analysis depend heavily on the capital market assumptions (CMAs) used. It may seem that running thousands of Monte Carlo simulations is “scientific,” showing what would happen to a portfolio under all possible future scenarios. But it is not. The results from Monte Carlo are entirely determined by the CMAs used.” “Probability of success, therefore, depends on how the Monte Carlo analysis is set up… But it is not all possible future values, or even a completely “random” selection of future values. The set of future values evaluated is determined entirely by the assumptions fed into the analysis.” “[This] survey provides the average CMAs across investment firms, as well as the variability in assumptions used by different investment firms.” www.advisorperspectives.com/articles/2023/01/10/the-dangers-of-monte-carlo-simulationsMustang’s Comment: The article shows the difference in capital market assumptions for 40 investment firms surveyed. Stock returns ranged from 5.15-10.63%. Bond returns ranged from 2.42-5.82%. Prediction of the future is clear as mud. Two Morningstar analysts, Chistrine Benz and John Rickenthaler in 2021 found that the initial withdrawal rate needed to be 3.3% with a 90% probability of success for the portfolio to last 30 years. This was published and quoted by many writers. www.morningstar.com/retirement/whats-safe-retirement-spending-rate-decades-aheadMichael Kitces wrote that the Morningstar study projected an average return of 3.5% for the 30 year period. That means in the simulations half of the scenarios had returns less than 3.5%. According to Shiller’s historical market data (1871-2021) the average return is 6.0%, the bottom quartile was 4.76%. Morningstar’s forecast of 3.5% would be far worse than the worst economy in history. “Morningstar’s choice to focus on (historically low) 30-year returns for its analysis disregards the evidence of what really drives safe withdrawal rates, which is the sequence of returns.” www.kitces.com/blog/4-percent-rule-bengen-morningstar-report-the-state-of-retirement-income-safe-withdrawal-rates/The market peaked in December of 2021. In December 2022 Benz and Rikenthaler updated their study. They said, “equity valuations declined and cash and bond yields have increased.” With updated computer inputs they said an initial rate of 3.8% is safe with a 90% probability of success. www.morningstar.com/retirement/whats-safe-withdrawal-rate-todayMustang’s Comments: One year later they changed the safe withdrawal rate to 4.0%. How confusing can that be for retirees? Do they use 3.3%. 3.8% or 4.0%? I did compare Morningstar’s computer simulations to Wade Pfau’s update to the Trinity Study which used historical data from 1926-2017. Here is the comparison for a portfolio that is 50% stock and 50% bonds. www.forbes.com/sites/wadepfau/2018/01/16/the-trinity-study-and-portfolio-success-rates-updated-to-2018/?sh=761f336e6860
Morningstar Morningstar Morningstar Pfau’s 2017 Payout Period 2021 Study 2022 Study 2023 Study Update 15-years 6.4% 6.6% 6.7% 6% 20-years 4.9% 5.2% 5.4% 5% 30-years 3.3% 3.8% 4.0% 4% 40-years 2.8% 3.2% 3.4% 3% Bottom line: Do not take Monte Carlo simulation as absolute. They are subject to analyst’s bias – a particular view of the future that cannot be seen. If you want to evaluate your portfolio for withdrawals test it against the worst case scenario. One that shows the impact of sequence-of-returns not randomized returns.
|
|
|
Post by yogibearbull on Jan 16, 2024 19:53:39 GMT
Monte Carlo runs include thousands of random cases (within the parameter ranges used) that should cover SOR situations too.
But if historic data are used, then one should use the worst periods.
|
|
|
Post by Mustang on Jan 16, 2024 23:10:25 GMT
A SOR scenario might be included but I'm not sure how much of an impact it would have. From Kitces’ review ( www.kitces.com/blog/4-percent-rule-bengen-morningstar-report-the-state-of-retirement-income-safe-withdrawal-rates/ ): “… it’s important to note that the Morningstar study used Monte Carlo analyses to calculate their safe withdrawal rates, meaning that their ‘projected’ return is really the average of a distribution of returns used to run each scenario… Sequence of Returns matters more than average withdrawals for safe withdrawal rates.” “Morningstar’s method of calculating forward-looking safe withdrawal rates involved first estimating a 30-year average return and standard deviation value for a given investment portfolio, then using those inputs to run a Monte Carlo analysis for the 30-year period that simulated the amount and sequence of returns for each scenario (while setting a constant rate of inflation to determine portfolio withdrawals over the 30-year time horizon).” “The problem with this method, however, is that Monte Carlo analysis generates a random sequence of returns for each simulation, when, in reality, market returns are generally not random from one year to the next. Instead, markets have historically followed secular bull and bear market cycles, typically lasting 10-to-20 years each, during which average returns performed higher (in bull market cycles) or lower (in bear market cycles) than their overall historical averages… “ “For retirees, therefore, it is not a matter of whether they will experience a period of below-average returns during retirement – for such periods will occur in almost any retirement horizon – but where in the succession of bull and bear market cycles their retirement date takes place that will have the most impact on their safe withdrawal rate.” “Morningstar’s approach toward re-examining the 4% rule – projecting safe withdrawal rates based on drastically reduced 30-year return projections – ignores the way markets have actually behaved in the past, focusing only on reduced 30-year returns (which usually encompass multiple up-and-down market cycles) rather than on the crucial first 10 to 15 years of the retirement horizon when sequence of return risk is most relevant… “ I’m still learning but this sounds to me like Monte Carlo’s random returns do not provide for sequence of return risk. If I read this correctly the 10,000 random scenarios are simply averaged together. This also caught my attention: ignores the way markets have actually behaved in the past. It may be the only way to test for SOR is using historical data.
|
|
|
Post by archer on Jan 17, 2024 0:33:46 GMT
What interested me in the possibility of backtesting is that PV allows us to run MC simulations on tickers, whereas most only run for asset allocations. Backtesting for asset allocation has two hurdles. One is that by nature, MC simulations only offer probabilities ranging from a low % to a high % of likelihood. This limitation in inherent in any MC simulation. It will show us that there is a 50% probability of things being average, which is pretty obvious to most of us, and decreasing odds of results being toward the extremes. The other limitation is that running a simulation for asset classes is using a vague data set. No distinction is made between a PF of the best funds in each asset class vs the worst. However for tickers we can be more specific. Since PV allows us to enter tickers, I compared 60/40 VFINX/VBMFX to PRWCX. In rolling periods PRWCX has done much better than the Vs, ands sure enough the MC simulations predict a much better result. I guess this shows that while past performance is not a guarantee, its better than nothing. Who wants to bet on a horse with a bad record?
In both simulations historical performance was used and thus gave different probabilities that coincided with real life, so it is onlyh telling us what we already knew. If the comparison above was ran prior to 2000, the historical data would have been different, and probably would not have favored PRWCX as strongly. I'm guessing PRWCX actually performed more toward the 90 percentile had the simulation been run many years ago.
|
|
|
Post by retiredat48 on Jan 17, 2024 2:07:04 GMT
archer,...in mathematically-inclined poster backtesting, the Bogleheads.org forum people concluded the sweet spot of safety was between 85/15 and 15/85 portfolios, before significant deterioration occurred. R48
|
|
|
Post by bb2 on Jan 17, 2024 2:21:45 GMT
Curmugeonly wary of the biz, here, again. IMO, monte carlo is another advisor tool to help sell. Just have a bunch of money and you'll be fine.
|
|
|
Post by fritzo489 on Jan 17, 2024 2:59:33 GMT
" Who wants to bet on a horse with a bad record? " 2022 Kentucky Derby winner Rich Strike. Paid $163.60 to WIN
|
|
|
Post by FD1000 on Jan 17, 2024 20:27:42 GMT
What interested me in the possibility of backtesting is that PV allows us to run MC simulations on tickers, whereas most only run for asset allocations. Backtesting for asset allocation has two hurdles. One is that by nature, MC simulations only offer probabilities ranging from a low % to a high % of likelihood. This limitation in inherent in any MC simulation. It will show us that there is a 50% probability of things being average, which is pretty obvious to most of us, and decreasing odds of results being toward the extremes. The other limitation is that running a simulation for asset classes is using a vague data set. No distinction is made between a PF of the best funds in each asset class vs the worst. However for tickers we can be more specific. Since PV allows us to enter tickers, I compared 60/40 VFINX/VBMFX to PRWCX. In rolling periods PRWCX has done much better than the Vs, ands sure enough the MC simulations predict a much better result. I guess this shows that while past performance is not a guarantee, its better than nothing. Who wants to bet on a horse with a bad record? In both simulations historical performance was used and thus gave different probabilities that coincided with real life, so it is onlyh telling us what we already knew. If the comparison above was ran prior to 2000, the historical data would have been different, and probably would not have favored PRWCX as strongly. I'm guessing PRWCX actually performed more toward the 90 percentile had the simulation been run many years ago. Yep, PRWCX is a rare bird and the type of fund I like and have used for years. Many years ago, I had several PMs with a retired poster who asked for an opinion if he can invests at 70% in PRWCX. After I was sure he knows the fund, the risk/reward + generic markets ideas, I said "go as long as Giroux is the manager". This is not a bad idea to find 3-5 flexible good allocation funds to do most of the heavy lifting.
|
|
|
Post by Mustang on Jan 17, 2024 20:42:27 GMT
What interested me in the possibility of backtesting is that PV allows us to run MC simulations on tickers, whereas most only run for asset allocations. Backtesting for asset allocation has two hurdles. One is that by nature, MC simulations only offer probabilities ranging from a low % to a high % of likelihood. This limitation in inherent in any MC simulation. It will show us that there is a 50% probability of things being average, which is pretty obvious to most of us, and decreasing odds of results being toward the extremes. The other limitation is that running a simulation for asset classes is using a vague data set. No distinction is made between a PF of the best funds in each asset class vs the worst. However for tickers we can be more specific. Since PV allows us to enter tickers, I compared 60/40 VFINX/VBMFX to PRWCX. In rolling periods PRWCX has done much better than the Vs, ands sure enough the MC simulations predict a much better result. I guess this shows that while past performance is not a guarantee, its better than nothing. Who wants to bet on a horse with a bad record? In both simulations historical performance was used and thus gave different probabilities that coincided with real life, so it is onlyh telling us what we already knew. If the comparison above was ran prior to 2000, the historical data would have been different, and probably would not have favored PRWCX as strongly. I'm guessing PRWCX actually performed more toward the 90 percentile had the simulation been run many years ago. I'm sorry but I'm a little confused. I didn't know Portfolio Visualizer did Monte Carlo simulations. Backtesting uses actual historical data not predictions of the future. Where do you enter predicted rates (both returns and inflation) and standard deviations?
|
|
|
Post by yogibearbull on Jan 17, 2024 20:51:01 GMT
|
|
|
Post by Mustang on Jan 17, 2024 21:21:48 GMT
Interesting. Thanks for the link.
Edit: For a quick test I used Wellington and Wellesley's historical data. $1M portfolio, 50/50 allocation, 4% Rule withdrawals. Even though the two fund have only lost money once two years in a row (73 & 74) I put the worst 10 years first (Testing for SOR risk was not mentioned by Morningstar concerning their simulation or Kitces during his review). This scenario is highly unlikely. After the first 10 years the portfolio dropped to just below half its starting value at the 50 percentile. They then ran out of money after 26 years.
Setting it to a more realistic worst 2 years first all except the bottom 10 percentile made it to 30 years. Overall probability of success was 84%. Considering all of the uncertainty surrounding predicting the future I don't need a 90% probability of success. Kitces once said that a 50% probability of success didn't mean a 50% probability of failure. It meant there was a 50% probability that changes would be necessary.
At the 50 percentile PV says a 5% initial withdrawal was possible. This pretty much matches what the Trinity Study found in 1998. It said that 17 of the 18 failures using a 5% initial withdrawal rate occurred during the stagflation years of double digit inflation. I personally would stick to 4% and increase the amount withdrawn should the portfolio do better than expected.
Again, thanks for the link.
|
|
|
Post by archer on Jan 17, 2024 23:07:51 GMT
What interested me in the possibility of backtesting is that PV allows us to run MC simulations on tickers, whereas most only run for asset allocations. Backtesting for asset allocation has two hurdles. One is that by nature, MC simulations only offer probabilities ranging from a low % to a high % of likelihood. This limitation in inherent in any MC simulation. It will show us that there is a 50% probability of things being average, which is pretty obvious to most of us, and decreasing odds of results being toward the extremes. The other limitation is that running a simulation for asset classes is using a vague data set. No distinction is made between a PF of the best funds in each asset class vs the worst. However for tickers we can be more specific. Since PV allows us to enter tickers, I compared 60/40 VFINX/VBMFX to PRWCX. In rolling periods PRWCX has done much better than the Vs, ands sure enough the MC simulations predict a much better result. I guess this shows that while past performance is not a guarantee, its better than nothing. Who wants to bet on a horse with a bad record? In both simulations historical performance was used and thus gave different probabilities that coincided with real life, so it is onlyh telling us what we already knew. If the comparison above was ran prior to 2000, the historical data would have been different, and probably would not have favored PRWCX as strongly. I'm guessing PRWCX actually performed more toward the 90 percentile had the simulation been run many years ago. I'm sorry but I'm a little confused. I didn't know Portfolio Visualizer did Monte Carlo simulations. Backtesting uses actual historical data not predictions of the future. Where do you enter predicted rates (both returns and inflation) and standard deviations? On the link that Yogi linked, there is a drop down menu for "Simulation Model". One of the possible models is "Forecasted Returns". To do use this model, you would have to as you say, specify returns and SD. However, the page doesn't provide a means of doing so. PV support is good about answering questions, but I don't have reason enough to take their time at this point. Testing for MC accuracy would be a matter of taking say 1000 simulations done say 30 years ago, and seeing how well they forecasted the future. After thinking about it more, there is no reason to do such a test. MC runs 10,000 scenarios of return sequence and provides results. Regardless of the actual results that transpire, the odds were what they were.
|
|
|
Post by Mustang on Jan 18, 2024 3:17:35 GMT
I was curious so I decided to try to compare Monte Carlo results to real world results. I ran a backtest on a 50/50 Wellington/Wellesley portfolio. 2000 start date, 4% initial withdrawal, ending 24 years later in 2023. Starting balance was $10,000, ending balance was $24,500. This was a stress test because the SP500 lost value in each of the first three years. www.portfoliovisualizer.com/backtest-portfolio?s=y&sl=6ZLJcN17RqeQCtGeIhUPnUSame setup for the Monte Carlo simulation using the 2000-2023 data. If I set it up correctly this is exactly the same data as the backtest: Again $10,000, 4% initial withdrawal, ending 24 years later (the chart went 25 years but 24 was visible). For sequence of returns the worst years were the first three. www.portfoliovisualizer.com/monte-carlo-simulation?s=y&sl=3ISeXCIWEvWmTReyBCeWinWe know with 100% certainty that backtesting the ending balance was $24,500. 10,000 simulations put the probability of success for 24 years at 87.2%. The closest ending balance was for the 75 percentile ($23,200). That would mean that using the same data 25% of the 10,000 scenarios had ending balances higher than the actual ending balance. Up to 10% of the scenarios didn't make it to 24 years. For the 75th percentile the safe withdrawal rate was calculated to be 5.89% when we know the actual withdrawal rate used was 4%. Just for grins I went back to the Backtest and changed the initial withdrawal to 5.89% ($589). The ending balanced dropped to $10,300. www.portfoliovisualizer.com/backtest-portfolio?s=y&sl=3sTUbHY2pQSpF5HUlcRJpIf I set it up properly, using the same data set (2000-2023) Monte Carlo simulations said a 5.89% withdrawal would have an ending balance of $23,200 at the 75th percentile when in the real world using a 5.89% withdrawal rate left an ending balance less than half of that ($10,300). Interesting results. I think I like using historical data better. It's not 10,000 simulations but it is real world.
P.S. Did I set up the Monte Carlo simulations correctly?
|
|
|
Post by archer on Jan 18, 2024 6:39:53 GMT
Mustang, Your comparison is exactly what I was thinking of, and shows the difference between what the odds were and how it panned out. I'm sure several of the 10,000 scenarios resulted the same % return to the 2nd decimal. But, to get a sense of how reliable the MC is, your comparison would need to be done to many PFs and the results averaged. It isn't clear to me if the 5.89% annual is based on running PF balance or if it represents the initial withdrawal which is then maintained inflation adjusted, (as in Bengen's 4% rule). I like using historical data too, and I always like to use the 2000 start year. Really though, all 10,000 scarios used in the simulations were in fact just as possible as the real historical results. I think using both is valuable in that they provide additional reference. In the end we still need to decide how safe is safe, and how much risk we want to take. We can see the worst and best possible outcomes given the SD, but even the SD can change. IOW the data set for VWINX and VWELX going forward could be much different than the past if the market becomes more or less volatile.
|
|