Forum

Policies – please read!

  • Everyone can read posts, but to start a new topic or add a reply you’d need to register. To create an ID, please verify your email address. You will get a link at that email address and you can then pick a password.
  • Be courteous to others. Treat others like you like to be treated. Discuss the issues. No ad-hominem attacks!
  • It’s OK to include external links if they are relevant. But please avoid spamming! Please no affiliate links.
  • Before starting a new topic, please check if that question/topic has been discussed before already and add to that discussion instead of starting a new topic.
  • If you’ve written a cool blog post that you want to share with others please post this in the “Self-Promo” category only!
  • Please read the usual Disclaimers and the Privacy Policy!
  • The forum policies may be amended in the future!
Notifications
Clear all

Algorithmic Glide Paths or AA Changes

Page 1 / 2

Chris B
Posts: 18
Topic starter
(@chris-b)
Active Member
Joined: 1 year ago

The issue we face is simply a tradeoff between (a) having too conservative an AA to last 30-40 years, and (b) avoiding damage from a SORR event in the first decade of early retirement. According to the chart at https://money.com/stock-market-correction-chart/  a 20% or greater correction happens on average every 4 years, and 30% or greater happens on average every 9 years. We should know it's coming. 

What if instead of setting a static AA or planning a glide path that is indifferent to the presence or absence of an actual SORR event, we instead start retirement with a very conservative AA, like 50/50 or 40/60, and wait for it? Then, when a pre-determined SORR threshold is reached, such as for example a 30% drop from the last high in the S&P500, we switch to an aggressive AA like 90/10 or 85/15 and hold that AA for the remainder of our retirement. 

The idea is we want to position ourselves to minimize damage from the eventual SORR event, but once the SORR event is actually occurring, it has historically been the safest time to pivot to an equities-heavy AA. This is a refinement of the glide path idea because the AA changes in response to an actual event, not a schedule. The AA changes once in response to information available at that time, and we favorably adapt to the often-fatal first SORR event of early retirement. 

-----

Imaginary example sequence of events:

1) Begin a 40 year retirement in January 2022 with 25x and a 40/60 stock/cash allocation. Our IPS says switch to a 90/10 AA if the S&P 500 drops 30% from its most recent high. This is our trigger event. (fill in your own preferred details or pick a mix of cash, bonds, and gold)

2) In September 2024, the stock market drops 30% but our portfolio is only down (40% * 30% =) 12%. We switch to a 90/10 AA as our IPS requires.

3) We didn't time it perfectly. The stock market has another 10% drop to go, costing our portfolio another 9% (total damage now -21%). We are still much better off than we would have been had we started with a 90/10 AA and suffered a 36% loss. We are now more aggressively allocated than if we were on a 10 year glide path though.

4) In June 2028 the market again reaches its previous highs. We took most of the downswing with a conservative AA and rode the recovery with an aggressive AA. Our portfolio has grown and we successfully dodged that critical first SORR event of early retirement. Because 40% drops are usually followed by long bull markets, we have a decade of above-average returns ahead of us which we capture in our aggressive AA, outrunning SORR events in the distant future.

------

I think this idea is worth looking into. Glidepaths allow for slightly higher WRs because they sometimes approximate the process I'm describing, slightly reducing damage from early SORR events and going slightly more aggressive as they occur. By tying the AA change directly to the actual event we are worried about, I hypothesize an algorithmic AA approach could save the 4% rule. 

ERN's SWR spreadsheet is not set up to simulate algorithmic AA changes like this, but if it was we could optimize the variables like starting AA, trigger size, and ending AA. I gave it a try for a couple of hours, but reverse-engineering other people's excel formulas is hard! Someone with more familiarity might knock it out much faster. Any takers?

15 Replies
Chris B
Posts: 18
Topic starter
(@chris-b)
Active Member
Joined: 1 year ago

One risk lies in the possibility that we set our trigger event too low. For example, if we switch from conservative to aggressive AA in response to a routine 15% correction, we aren't dodging much risk in terms of retirement-killing bear markets. Another problem would be if the stock market behaves in an ahistorical way and somehow does not have a correction the size of our trigger event for decades. Such a happy scenario would probably be a cohort where a hyper-conservative portfolio would do fine anyway. Last, there might be "double tap" scenarios where a correction triggers us to go aggressive and then we are hit with a second, bigger correction immediately thereafter. Between these risks is probably a sweet spot where one's trigger is big enough to mitigate SORR but small enough to mitigate the risk of sitting in a conservative portfolio too long. Then there is behavioral risk. Increasing one's AA while people are dying in the streets from Ebola, while China is invading its neighbors, or while the US government is being overthrown would be hard and one might say "how about a 50% drop instead of 30%? Sounds reasonable given the crazy circumstances." Then you miss the bottom and your retirement fails. 

Reply
earlyretirementnow.com
Posts: 294
(@earlyretirementnowcom)
Member
Joined: 6 years ago

I generally like the idea. I haven't simulated that yet because there are too many bells and whistles and different ways to model this. As you noted in your second comment, you might move too soon after a 15% drop, and then if the world goes to hell you wipe out your 90/10 portfolio too fast. Or you set the target drop too pessimistically and miss the turning point.

There's also a behavioral component: Are we sure that retirees from the January 2020 cohort had the nerve to move to 90/10 at the bottom of the bear market on March 23, 2020. If they did that they would have done phenomenally well. 

So, yes, sounds like a nice idea in theory, but the practice part is not 100% clear. But certainly, a good idea to research, so thanks for the suggestion. 🙂

Reply
Chris B
Posts: 18
Topic starter
(@chris-b)
Active Member
Joined: 1 year ago

I've been trying to hack the SWR spreadsheet to model this algorithmic behavior. I've got the variables in place, got the trigger detection columns in place, but the point where I get stuck is how to model independent behavior for each specific cohort. The spreadsheet's existing infrastructure is oriented toward averaging the returns of one AA rather than listing the specific experience of cohort after cohort as they change AAs.

E.g. if someone retired in 1985, they probably go aggressive in 1987, but someone retiring in 1988 sticks with conservative - possibly for a long time. 

I'm racking my brain trying to come up with a simpler solution than creating a monthly cohort by monthly cohort specific simulation that adds up the cumulative returns, net of withdraws, for the conservative AA until the trigger occurs, and then switching to the aggressive AA. Even that would involve some formulas I'm unfamiliar with. Any alternative ideas?

Reply
4 Replies
earlyretirementnow.com
(@earlyretirementnowcom)
Joined: 6 years ago

Member
Posts: 294

@chris-b Any kind of algo trading, like momentum, can be done through the "Custom Series". Did you try that?

Reply
Chris B
(@chris-b)
Joined: 1 year ago

Active Member
Posts: 18

Thanks for the lead. I understand the custom series to be a place users can plug in returns for alternative assets, am I right? I would need to come up with returns data to overwrite these cells. I was originally considering testing the following rule:

Rule 1:

A retiree shall start with a conservative AA and then permanently switch to an aggressive AA after the S&P500 has declined X% from its most recent high.

The X% decline would be somewhere around the 20%+ range where the 4% rule survived 100% of the time per the spreadsheet's "SWR to target Failure Rates, Conditional on S&P500 Drawdown" table. The idea is if a retiree could hold a safe portfolio until such an event happened, as they do every few years, then the retiree could essentially begin their retirement anew at the time of the correction with a higher SWR. Thus waiting a couple of years for a big correction post-retirement would be the most important thing. However, the risk of such an algo is fatally drawing down the cash and bond heavy portfolio during the potentially long period of time that passes until that SORR event occurs. This is an especially big problem now that real interest rates are negative! You could have a cash drawdown portfolio for years.

I can't figure out a way to model the rule above without creating a returns sequence for each and every cohort. E.g. If I retire in 2007, and there is a SORR event in 2008, I switch to aggressive in 2009. If my friend retires in mid-2009, he starts with a conservative AA, even though I'm running an aggressive AA. Can you figure out how to do this with the custom series?

A much simpler rule to test using the existing infrastructure of the spreadsheet and the custom series fields would be:

Rule 2:

A retiree shall at all times look back X months and determine if a S&P500 drawdown of Y% from its most recent high has occurred during that time. If yes, hold or switch to an aggressive AA. If no, hold or switch to a conservative AA.

Both rules have their appeals and drawbacks. I can imagine scenarios where they would yield different outcomes for different cohorts.

The 1st rule would layer the benefits of a conditional AA strategy with the proven benefits of a glidepath on steroids, and might better survive several consecutive years of single digit negative returns that don't quite trigger the AA change. Then, because the change is permanent, most cohorts would probably see their aggressive portfolios go to the moon over the decades after they survive that critical first SORR event.

The 2nd is a lot easier to model, but might move a retiree in and out of AA's based on past market conditions which are no longer relevant - E.g. going aggressive upon retirement now even though 2020's 30% drawdown and attractive valuations have been more than erased. Then again, if we could dial in the optimal time span, this counter-cyclicity might work in its favor. It might be just the nudge to get a 1920's retiree into a conservative portfolio prior to the 1930s, or similar for a 1960's or 1990's retiree to the extent that helped. 

I might try to model the 2nd rule and see how it goes. I'm more interested in the 1st rule though. Feel free to do a post on these ideas if you want, because it could be months before I get around to this stuff, not being retired and all!

Reply
earlyretirementnow.com
(@earlyretirementnowcom)
Joined: 6 years ago

Member
Posts: 294

@chris-b If the rules depend on the S&P500 stats, then you can just simulate that, package it into one series and feed that into the custom return series.

I'd be interested in seeing the results.

Reply
andyg42
(@andyg42)
Joined: 5 months ago

Eminent Member
Posts: 27

@chris-b I've hacked the SWR sheet, the "Case Study" tab that has the glidepath for a given cohort retirement month in particular, to do something very much like you're suggesting.

 

I've actually done 3 things with it:

A. A simplified Guyton-Klinger-like rule set ("guardrails") that avoids the problems ERN wrote about to:

  1. Ratchet up spending whenever real portfolio value increases ["retire again"]
  2. Cut spending by 10% from the latest baseline if portfolio value down 35%+ the first 15 years, and
  3. Cut spending by an additional 10% if portfolio down by 50%+ in the first 20 years

 

B. A fairly simple calculation where the next month's stock allocation is a function of (the six month moving average of) how much stocks are down from their high

 

C. Potentially modify that above stock allocation to:

  1. Maintain momentum: if stocks close to their all-time high, don't reduce equity allocation from prior month
  2. Avoid catching a falling knife: Reducing equity allocation substantially (15 or 20 percentage points) when stocks fall 10% below the high and keeping it low for a few months past the most recent stocks trough
  3. Press equity bets for a long while after a trough: Don’t reduce equity allocation below prior if within 21 months of most recent trough unless stocks hit a new peak

 

I do all the above by adding columns to the hacked Case Study tab that calculate:

  • S&P max to date
  • Real S&P value
  • Real S&P max to date
  • % S&P down from real high
  • % S&P down from nominal high
  • Trough since last S&P (nominal) peak
  • Months since last trough
  • % portfolio down from its high

 

The results seem pretty impressive, when tested against the worst retirement dates since 1926, in terms of final values and relatively high SWRs / average spending rates. And even with fairly high initial SWRs, for a 30 year retirement most cohorts other than 1937 rarely have to cut spending even 10% for more than a handful of months, and almost never requiring a temporary 20% cut

 

I hope to do a Forum post with more details on this in the next day or two.

Unfortunately, the hacked sheet itself is not yet fit for consumption by others. Big ERN, as the creator of the original, could probably eventually figure it out, but he might be the only one!

 

Reply
Chris B
Posts: 18
Topic starter
(@chris-b)
Active Member
Joined: 1 year ago

OK, so I've adapted the spreadsheet to allow for simulations to explore hypothesis #2, which was

Rule 2:

A retiree shall at all times look back X months and determine if a S&P500 drawdown of Y% from its most recent high has occurred during that time. If yes, hold or switch to an aggressive AA. If no, hold or switch to a conservative AA.

I haven't done every iteration of the idea, but so far the results do not look good for rule #2. Changing one's AA from a defensive to an aggressive AA after a big drop in an attempt to catch the rebound almost always seems to worsen results. It doesn't even help to do the opposite, entering a positive number as the lookback period to chase momentum.

Perhaps this is because the bear markets we are trying to avoid are of the many-years-long variety, and going aggressive in the middle of such a downturn results in cuts from falling knives. You can improve on the failure probabilities of highly defensive portfolios (e.g. 40/60) but the failure probabilities are still worse than I could stand, and this result is probably due to time in an aggressive AA, not the change itself.

OTOH, maybe the algorithmic models trade one sort of risk for another. E.g. If I retire right before a long bull market with a defensive AA, maybe I fail because I miss the gains and draw down a relatively flat portfolio. I'd have to dive into the specific problem areas to understand the "why" further. 

Spreadsheet notes:

1) All edits are highlighted in yellow for your convenience. I was able to reuse almost all of the spreadsheet's existing infrastructure. This was the "easy" hypothesis to test. 

2) I still need to explore what happens when larger numbers are entered as the lookback duration and the aggressive AA duration. The formulas don't break because these only select the AA in StockBondReturns, not expanding an array or anything. The defensive AA is selected by default in the Portfolio return column, so the effect of an error would be to stay defensive longer than a hand-tabulation would yield.  

3) The longer the lookback period, the more of the 1870's get defaulted to a defensive portfolio. One could in theory fix this by reworking the SWR formulas to start in the 1880's.

4) I could drop this into a google sheet for security if you prefer.

Overall, I'd like to know if you see anything wrong with the way I designed this experiment? I was hopeful that retirees could "buy the dip" or at least benefit from the scenario we think about with SORR and bond tents. Unless I've made a conceptual or spreadsheet error, it doesn't look like Rule #2 can save the 4% rule.

Reply
3 Replies
earlyretirementnow.com
(@earlyretirementnowcom)
Joined: 6 years ago

Member
Posts: 294

@chris-b Not sure what's wrong, but all of the calculations are messed up. I see only "#REF!" as outputs.

Reply
andyg42
(@andyg42)
Joined: 5 months ago

Eminent Member
Posts: 27

@chris-b after staring at the different worst case scenarios for a while, here's what i did to combat the problem you found:

1) go from normal/aggressive to conservative AA whenever stocks down 10% from high

2) stay at conservative AA until 4 months past the trough (as measured since the most recent peak), so as not to try to "catch a falling knife"), then switch to hyper-aggressive AA

3) maintain that hyper-aggressive AA for either 21 months or until stocks hit a new high.

 

I picked 21 months since I observed that there was always at leas 24 months after a severe trough until another big downturn started.

 

I just started a new topic describing what I did in more detail. I hope to post the hacked spreadsheet soon

Reply
andyg42
(@andyg42)
Joined: 5 months ago

Eminent Member
Posts: 27

Hacked sheet posted here:

https://earlyretirementnow.com/Forum/swr/hey-i-just-saved-the-4-rule-for-60-year-retirements-ratchets-guardrails-variable-equity-allocation/paged/3/#post-1861

Once I realized I was doing a market timing AA system rather than just a glidepath replacement, I changed “conservative” allocation to sell all equities. And frequently the “press equities” is 100% for stretches.  The results are substantially better SWRs.

I also include the Meb Faber simple MMA there as well. It has only a single signal, so by definition stocks are either 100% or zero.  The results are much better than glidepath, but not as good as my “home brewed” method

Reply
Page 1 / 2
Share: