Interesting FireCalc Results

Discuss all general (i.e. non-personal) investing questions and issues, investing news, and theory.
Post Reply
Topic Author
OldSport
Posts: 389
Joined: Tue Jul 25, 2017 7:01 pm

Interesting FireCalc Results

Post by OldSport » Fri Jan 11, 2019 8:11 pm

I got the best FireCalc results supporting a 4% SWR at 100% success for a 30 year period getting the following results:

70-80% stocks
20-30% bonds

Stock portion equally divided: 25% S&P 500, 25% Large Cap Value, 25% Small Blend, 25% Small Value.

Increasing the stocks above ~82% quickly reduced success rate and reducing below ~64% quickly reduced success rate. Having more than 50% small increased average gains but reduced success rate. Decreasing small and value tilt reduced the success rate as did increasing them.

Very interesting.

International was not an option, but would be very interesting in how introducing any portion (Total, EM, small cap) would impact success rate.

User avatar
David Jay
Posts: 6938
Joined: Mon Mar 30, 2015 5:54 am
Location: Michigan

Re: Interesting FireCalc Results

Post by David Jay » Fri Jan 11, 2019 9:05 pm

This is in keeping with most of the research. The issue is that, from a behavioral standpoint, many people can’t handle 70-80% in stocks.
Prediction is very difficult, especially about the future - Niels Bohr | To get the "risk premium", you really do have to take the risk - nisiprius

bluquark
Posts: 693
Joined: Mon Oct 22, 2018 2:30 pm

Re: Interesting FireCalc Results

Post by bluquark » Fri Jan 11, 2019 9:18 pm

A major reason for that is that inflation slammed bonds in the 70s, and this was permanent damage. Equities have always recovered in a timely manner in the US so recessions/depressions never really threatened SWR success rates. It's a bit of an artifact of the country: I imagine Japanese-based backtesting (assuming domestic Japanese investments only) would favor a more bond-heavy AA.

Beliavsky
Posts: 932
Joined: Sun Jun 29, 2014 10:21 am

Re: Interesting FireCalc Results

Post by Beliavsky » Fri Jan 11, 2019 9:49 pm

OldSport wrote:
Fri Jan 11, 2019 8:11 pm
I got the best FireCalc results supporting a 4% SWR at 100% success for a 30 year period getting the following results:

70-80% stocks
20-30% bonds

Stock portion equally divided: 25% S&P 500, 25% Large Cap Value, 25% Small Blend, 25% Small Value.
We know that going forward, this portfolio does not have a 100% chance of success, since the past will not repeat exactly. A Monte Carlo simulation may give more realistic success rates.

rkhusky
Posts: 7029
Joined: Thu Aug 18, 2011 8:09 pm

Re: Interesting FireCalc Results

Post by rkhusky » Sat Jan 12, 2019 8:39 am

Beliavsky wrote:
Fri Jan 11, 2019 9:49 pm
We know that going forward, this portfolio does not have a 100% chance of success, since the past will not repeat exactly. A Monte Carlo simulation may give more realistic success rates.
Or not, depending on the specifics of the Monte Carlo simulation. Many of them do not adequately capture sequence of returns. And since the distribution of stock market returns is unknown and probably non-stationary, the particular statistical distribution chosen and how it varies over time will play a role in the results.

blahblahsunshine
Posts: 55
Joined: Sun Jan 14, 2018 8:11 pm

Re: Interesting FireCalc Results

Post by blahblahsunshine » Sat Jan 12, 2019 3:14 pm

rkhusky wrote:
Sat Jan 12, 2019 8:39 am
Beliavsky wrote:
Fri Jan 11, 2019 9:49 pm
We know that going forward, this portfolio does not have a 100% chance of success, since the past will not repeat exactly. A Monte Carlo simulation may give more realistic success rates.
Or not, depending on the specifics of the Monte Carlo simulation. Many of them do not adequately capture sequence of returns. And since the distribution of stock market returns is unknown and probably non-stationary, the particular statistical distribution chosen and how it varies over time will play a role in the results.
With Rkhusky on this one... I often see manipulation in the assumptions of monte carlo analysis. One of my favorite is personal capital that down tamps the basis of the model 1% (given where we are at in the econmic cycle as they say). There is also the matter of how MC treats each interation of the model independently from the one before it too. While the past isn't a perfect predictor of the future it is in fact a perfect descriptor of the past...and a very reasonable way (in my opinion) to think about the future....

User avatar
Tyler9000
Posts: 429
Joined: Fri Aug 21, 2015 11:57 am

Re: Interesting FireCalc Results

Post by Tyler9000 » Sat Jan 12, 2019 3:24 pm

OldSport wrote:
Fri Jan 11, 2019 8:11 pm
International was not an option, but would be very interesting in how introducing any portion (Total, EM, small cap) would impact success rate.
Try this: https://portfoliocharts.com/portfolio/withdrawal-rates/

The tradeoff for having more assets is that you have less years of data (read the FAQ for specific info on how that affects the numbers), but this should help you explore the topic in more depth.

2015
Posts: 2906
Joined: Mon Feb 10, 2014 2:32 pm

Re: Interesting FireCalc Results

Post by 2015 » Sat Jan 12, 2019 8:08 pm

David Jay wrote:
Fri Jan 11, 2019 9:05 pm
This is in keeping with most of the research. The issue is that, from a behavioral standpoint, many people can’t handle 70-80% in stocks.
This.
And many people think they can until life gets in the way.

afan
Posts: 4302
Joined: Sun Jul 25, 2010 4:01 pm

Re: Interesting FireCalc Results

Post by afan » Sun Jan 13, 2019 3:56 pm

blahblahsunshine wrote:
Sat Jan 12, 2019 3:14 pm
rkhusky wrote:
Sat Jan 12, 2019 8:39 am
Beliavsky wrote:
Fri Jan 11, 2019 9:49 pm
We know that going forward, this portfolio does not have a 100% chance of success, since the past will not repeat exactly. A Monte Carlo simulation may give more realistic success rates.
Or not, depending on the specifics of the Monte Carlo simulation. Many of them do not adequately capture sequence of returns. And since the distribution of stock market returns is unknown and probably non-stationary, the particular statistical distribution chosen and how it varies over time will play a role in the results.
With Rkhusky on this one... I often see manipulation in the assumptions of monte carlo analysis. One of my favorite is personal capital that down tamps the basis of the model 1% (given where we are at in the econmic cycle as they say). There is also the matter of how MC treats each interation of the model independently from the one before it too. While the past isn't a perfect predictor of the future it is in fact a perfect descriptor of the past...and a very reasonable way (in my opinion) to think about the future....
MC would not have to treat each iteration as independent, but that is a common approach.
The main point is that the major limitations of deterministic, historical and MC are not the depth of the analysis but the noisy data. Each requires some assumptions about future returns. Given all the errors in those assumptions, it can be a mistake to assume more precision in the results than the data support.

Whether the relative return differences between S&P 500, small and small value will persist in the future is anyone's guess. The relative returns vary widely with period as it is. For firecalc I just use S&P 500 for "stocks". That already requires an assumption of return for an asset class. Breaking stocks down into component classes and pretending I know the relative performance is making assumptions I don't need to make.

All these methods give one a general idea of a range of possible outcomes. When doing this, vary the inputs over the full plausible range (using your judgement as to what that range might be) and try to extract some conclusions. If you find that variations of 2% or 5% in asset allocation are driving your results then you are far into false precision territory.
We don't know how to beat the market on a risk-adjusted basis, and we don't know anyone that does know either | --Swedroe | We assume that markets are efficient, that prices are right | --Fama

Post Reply