Firecalc accuracy and future.

[clifp]

In light of some of things NW-Bound has discovered about FIRECalc discussed in this thread.

I am wondering if there any plans to update either the functionality or at least make a better description of how FIRECalc operates.

I am sure I am not the only one who expected that capital appreciation of bonds to be factored in FIRECalc runs. Perhaps even more concerning is if interest rates do in fact rise those with large bond allocation will see a decrease in the value of their bonds fund, even while the yields rise.

 

[Lsbcal]

NOTE: I am going to repeat the posting I made over in the thread mentioned in the OP here. This shows some of the data I took from FIRECalc and a very brief comparison to validate the problem.
------------------------ repeated comments below ----------------------

...(snip)...
I examined the spreadsheet downloaded from FIRECalc and saw the following. FIRECalc only applies the income or dividend of the bonds to its growth. FIRECalc does not consider the appreciation of the principal when the interest drops, and the bond yield follows it.
...

The more I thought about this one, the more I worried about what FIRECalc is doing with this data. The 3 bad scenarios I mentioned above would be greatly affected by not considering the interest rate rise/fall (so called capital returns).

So I did a test for years 2002 to 2011. I looked at a pure 5 year Treasury portfolio of $1M with no spending. FIRECalc gave a final value of $1,064,095 inflation adjusted. Then I looked at what $1M would have done in VFIUX (Vanguard 5 year Treasury admiral). The final inflation adjusted value was $1,457,944 (for total return, green area below). This is a huge difference. Then I just used the income return for VFIUX (blue area below). The result was much closer to FIRECalc $1,173,121 indicating FIRECalc does indeed ignore the capital returns.

Did I do this right? Do you guys agree that this is a problem?

Here is the relevant parts of the spreadsheet:

 

[ERD50]

...

Did I do this right? Do you guys agree that this is a problem?
...

I'm not so sure this is a problem. But it raises a question - What should FIRECalc assume?

IOW, what does it really mean to have money invested in 5 year treasuries for 30 years? If you hold them (purchased 6 times, every 5 years) to maturity, you see no valuation or interest rate change for 5 years.

So should it assume you buy them, hold for 5 years and then buy new issues? If so, the interest rate should remain the same for 5 years. Any rebalancing would mean trading and a change in rates if buying (or buy the same issue at current NAV), and selling should reflect the current NAV.

Or assume a 5 year ladder? Now that could be complicated for the first 5 years - do you start with 1-2-3-4-5 year issues, ans replace the 1's with 5's, etc?

It seems complicated, it isn't just the NAV of 5 year treasuries, it has to do with their maturity. I don't think there are any bond indexes to track going back far enough. Maybe one can be simulated (I guess the ladder does this?)

Or did I miss something?

-ERD50

 

[CaliforniaMan]

Here's the thing I've found with FIRECalc.
...
FIRECalc only applies the income or dividend of the bonds to its growth. FIRECalc does not consider the appreciation of the principal when the interest drops, and the bond yield follows it.
...

...
The more I thought about this one, the more I worried about what FIRECalc is doing with this data. The 3 bad scenarios I mentioned above would be greatly affected by not considering the interest rate rise/fall (so called capital returns).

So I did a test for years 2002 to 2011. I looked at a pure 5 year Treasury portfolio of $1M with no spending. FIRECalc gave a final value of $1,064,095 inflation adjusted. Then I looked at what $1M would have done in VFIUX (Vanguard 5 year Treasury admiral). The final inflation adjusted value was $1,457,944 (for total return, green area below). This is a huge difference. Then I just used the income return for VFIUX (blue area below). The result was much closer to FIRECalc $1,173,121 indicating FIRECalc does indeed ignore the capital returns.

Did I do this right? Do you guys agree that this is a problem?
...

This seems to me to be a HUGE problem! It must mean that the decline of principal is also not calculated when interest rates rise, bringing into question the whole re-balance modelling calculations in FIRECalc.

A while ago I was wondering what duration FIRECalc used to make its re-balance calculations, because of course you cannot estimate principal bond price changes without knowing the duration.

In fact I was thinking about posting a question here about that. I assumed that they must use some fixed duration, since I couldn't find an input for it. But then again, I am a novice FIRECalc user.

If such an obvious measurement as bond duration and the principal changes with interest rates are not modeled correctly in FIRECalc, it calls into question all of its projected SWR probabilities.

If what NW-Bound and Lsbcal have found are correct it seems to me that this is a major flaw.

 

[Gatordoc50]

I've never been a fan of firecalc as a retirement tool but rather as an entertainment tool. Back testing bonds seems foolish when the best predictor of future bond yields may be the present yield curve. Probably the best way to use firecalc is to separate out your equities and run it with 100% stocks and test your bond allocation without interest rate risk.

 

[clifp]

I'm not so sure this is a problem. But it raises a question - What should FIRECalc assume?

IOW, what does it really mean to have money invested in 5 year treasuries for 30 years? If you hold them (purchased 6 times, every 5 years) to maturity, you see no valuation or interest rate change for 5 years.

So should it assume you buy them, hold for 5 years and then buy new issues? If so, the interest rate should remain the same for 5 years. Any rebalancing would mean trading and a change in rates if buying (or buy the same issue at current NAV), and selling should reflect the current NAV.



Or did I miss something?

-ERD50

No I don't think you are missing much. I agree there is no easy solution. In large part because we don't have bond fund data going back very far. Most early Vanguard bond funds were started between 78-83 and the total bond index admiral fund not until 2001.

At minimum I think the results need warning prominently displayed.

CHANGES IN BOND VALUES DUE TO INTEREST RATE FLUCTUATIONS ARE NOT MODELED IN FIRECALC.

If you carefully read the description of how FIRECalc works it does describe it accurately.

Then, FIRECalc computes what would have happened to the remainder of your portfolio for that year, with market growth, dividends, investment expenses, and interest on your fixed income investments.

Although I'd say that changing the sentence.
"with market growth, dividends, investment expenses on your equity investments, and INTEREST ONLY on your fixed income investments" would be much clearer


Still I have been using FIRECalc for 7 years and never caught on. One of the significant benefits of having both bonds and stocks is the low correlation between the two asset classes. So when stock skyrocketed in 99 and interest rates rose. rebalancing at the beginning of 2000 (like I did) had me buy cheap bonds. This help greatly cushioned 2000-2002 bear market. Likewise I know many people were helped in 2008 and 2009 by rising value of their bonds.


A more complicated fix would model the value of the bond value based the duration of your fixed investments and the change in interest rates. So for example if you had 30 year Treasury bills we know that duration is about 15 years. If interest rates increase 1% next year the value will drop by 15% and you'll collect about 4% in interest leaving you with an 11% loss in your bond portfolio. Commercial paper probably a 5 year duration.

I disagree with GatorDoc. I think FIRECalc is the best tool precisely because uses historical date and not Monte Carlo simulation. Especially because well meaning and smart guy like Professor Pfau can't help but make predictions of the future returns for various asset classes.

History doesn't repeat itself but financial history absolutely rhymes.

 

[gcgang]

One of the significant benefits of having both bonds and stocks is the low correlation between the two asset classes.

I think FIRECalc is the best tool precisely because uses historical date and not Monte Carlo simulation. .

I think the above two statements are very dangerous.

While its true a Monte Carlo simulator can be dangerous in the hands of an inexperienced operator, in that instead of historical data, you can adjust returns and standard deviations to fit your outlook, if done well, it can be very illuminating. For example, if you wanted to be ultra conservative in your modeling, you could use lower equity returns than the last (dominated by USA) century, and could increase standard deviation to make sure your portfolio could withstand periods worse than what we've historically witnessed.

As far as stocks and bonds being negatively correlated, the below link shows that sometimes that is the case, and sometimes not. One of the drivers of the Great Bull market of the 1980's was the decline in interest rates. Both bonds and stocks made excellent returns. My guess is that if we mirror the 1980s, and interest rates increase significantly from current levels, bringing bond prices down, equities will also suffer.
Correlation Between Stocks And Bonds - Business Insider

None of these tools do a good job of modeling what I think is one of the greatest risks we FIREs face, that of a short, or worse an extended, period of very high inflation.

 

[ERD50]

...
While its true a Monte Carlo simulator can be dangerous in the hands of an inexperienced operator, in that instead of historical data, you can adjust returns and standard deviations to fit your outlook, if done well, it can be very illuminating. For example, if you wanted to be ultra conservative in your modeling, you could use lower equity returns than the last (dominated by USA) century, and could increase standard deviation to make sure your portfolio could withstand periods worse than what we've historically witnessed. ...

OK, but then how do I know if the conditions I input were truly conservative enough?

Well, I could compare them to the worst we've had in history. Hmmm, then why not just use the historical data?

And if you want be more conservative than history, that's easy, just add whatever buffer factor makes you comfortable.

edit/add - to clifp's earlier post - yes, I think a disclaimer would be a good thing.

-ERD50

 

[NW-Bound]

...
So I did a test for years 2002 to 2011. I looked at a pure 5 year Treasury portfolio of $1M with no spending. FIRECalc gave a final value of $1,064,095 inflation adjusted. Then I looked at what $1M would have done in VFIUX (Vanguard 5 year Treasury admiral). The final inflation adjusted value was $1,457,944 (for total return, green area below). This is a huge difference...

I observed a similar big difference in a test case, as reported in the other thread. See below.

...
But here's the most interesting point. At the end of 2011, FIRECalc shows that the 50/50 blend of Total Stock/Long Treas portfolio balance would be $714K in nominal dollars (see earlier post), while my spreadsheet shows $1183K. That's a big difference!

The answer to the $469K question is next.

I'm not so sure this is a problem. But it raises a question - What should FIRECalc assume?

IOW, what does it really mean to have money invested in 5 year treasuries for 30 years? If you hold them (purchased 6 times, every 5 years) to maturity, you see no valuation or interest rate change for 5 years.

So should it assume you buy them, hold for 5 years and then buy new issues? If so, the interest rate should remain the same for 5 years. Any rebalancing would mean trading and a change in rates if buying (or buy the same issue at current NAV), and selling should reflect the current NAV.

Or assume a 5 year ladder? Now that could be complicated for the first 5 years - do you start with 1-2-3-4-5 year issues, ans replace the 1's with 5's, etc?

It seems complicated, it isn't just the NAV of 5 year treasuries, it has to do with their maturity. I don't think there are any bond indexes to track going back far enough. Maybe one can be simulated (I guess the ladder does this?)

Or did I miss something?

-ERD50

...
A while ago I was wondering what duration FIRECalc used to make its re-balance calculations, because of course you cannot estimate principal bond price changes without knowing the duration.

In fact I was thinking about posting a question here about that. I assumed that they must use some fixed duration, since I couldn't find an input for it. But then again, I am a novice FIRECalc user.
...

FIRECalc offers the choice of 30-yr and 5-yr Treasury. However, the problem is that bonds are not like stocks. With the latter, all shares are fungible, at least the ones of the same class. Not so with bonds. Individual bonds have different coupon rates, different maturity dates, so how do we model them all?

I have never bought individual bonds, except for I-bonds. And as I-bonds cannot be resold on the aftermarket, they are a lot simpler. When it comes to corporate or government bonds, I buy an MF. A 30-yr bond MF like VUSTX shows a duration of around 15 years. I guess it has a ladder of bonds, hence the average duration is 1/2 of that of the newest bond in its portfolio. The yield is the average of all the individual bonds in the portfolio.

So, modeling individual bonds is going to be tough, if we want to mark them to market value. With the MF, perhaps there are existing bond indices that can be used. However, these indices may not go back enough to study historical portfolio performance.

I guess FIRECalc uses the published historical coupon rates to compute the bond income. This however does not match the income of either the individual bond buyer, nor the MF share holder. And it misses the market value of the principal, which is important for a portfolio balancer.

What to do?

 

[clifp]

I think the above two statements are very dangerous.

While its true a Monte Carlo simulator can be dangerous in the hands of an inexperienced operator, in that instead of historical data, you can adjust returns and standard deviations to fit your outlook, if done well, it can be very illuminating. For example, if you wanted to be ultra conservative in your modeling, you could use lower equity returns than the last (dominated by USA) century, and could increase standard deviation to make sure your portfolio could withstand periods worse than what we've historically witnessed.

As far as stocks and bonds being negatively correlated, the below link shows that sometimes that is the case, and sometimes not. One of the drivers of the Great Bull market of the 1980's was the decline in interest rates. Both bonds and stocks made excellent returns. My guess is that if we mirror the 1980s, and interest rates increase significantly from current levels, bringing bond prices down, equities will also suffer.
Correlation Between Stocks And Bonds - Business Insider

None of these tools do a good job of modeling what I think is one of the greatest risks we FIREs face, that of a short, or worse an extended, period of very high inflation.

Taken your 2nd point first, if you look at your chart it looks like about a dozen years were bonds had a high correlation about a dozen year out of the last 60,and had a negative correlation in even more years. Bernstein in his e-book book about correlation quotes .05% correlation between long bonds and stocks and .03% been T-bills stocks. That is close enough to 0 for me. As Bernstein points it is increasingly hard to find asset that are not correlated, small cap, large value, Europe, Emerging markets, and the S&P all have move more or less in lock step for the last decade making achieving diversification very difficult. So really only gold, bonds, and real estate provide diversified alternative assets which is why I think it is important that FIRECalc models bonds/fixed income better.

Like ERD I am willing to assume that I am not unlucky enough to be faced with inflation, market performance etc that are worse than the last hundred plus year. I also know that I am not nearly smart enough to make prediction of the long term returns of a bunch of asset classes.

So I find Monto Carlo simulators to be really dangerous forecasting tool, perhaps especially in the hand of experienced operators.

To pick on Dr. Pfau again, in his Monte Carlo simulator observed that US equity had better returns than the rest of the world so he lowered equity returns by at least 2%, which coupled with the Monto Carlo flaws, had him suggest withdrawal rate down in the 2.5% range and 30-40% equity position.

Another expert I think it was Speigel did a study that showed among English speaking/Commonwealth countries US equity performance is actually fairly average over the last 100 years of so.

In my opinion there are structural reason the US, Australia, Canada etc. stock market returns have been historically so much better. I also am sure that financial market over the last 30 odd years have become much more like the US rather than vice versa, so it seem logical that rest of the world equity returns will be more in line with the US than vice versa. But of course I could be dead wrong.

But the really dangerous thing about Monto Carlo simulators is they
assume that annual performance of the stock market is independent of the previous years, when clearly it is not. Plenty of people knew the stock market was really over valued in 1999. Why because the NASDAQ was up more than 400% in 5 years. Now I didn't know that 2000 would the year of the crash but I was very confident that we were far more likely to have bad bear market in the next couple of years than normal. Likewise plenty of folks on the board knew in 2009 that a bull market was far more likely than another bear market, why because the market was down almost 40% in 2008.

The reason I think Monto Carlo is dangerous is because the operator bad assumptions can lead to dumb behavior. For instead if the $1 million retiree had followed Dr Pfau advice and switched from a 60/40 to 30/70 AA based on his model in the last couple of year they are out more than $100,000 and perhaps just as importantly were spending at a level of only 2/3 what they could have.

 

[audreyh1]

I'm not so sure this is a problem. But it raises a question - What should FIRECalc assume?

IOW, what does it really mean to have money invested in 5 year treasuries for 30 years? If you hold them (purchased 6 times, every 5 years) to maturity, you see no valuation or interest rate change for 5 years.

So should it assume you buy them, hold for 5 years and then buy new issues? If so, the interest rate should remain the same for 5 years. Any rebalancing would mean trading and a change in rates if buying (or buy the same issue at current NAV), and selling should reflect the current NAV.

Or assume a 5 year ladder? Now that could be complicated for the first 5 years - do you start with 1-2-3-4-5 year issues, ans replace the 1's with 5's, etc?

It seems complicated, it isn't just the NAV of 5 year treasuries, it has to do with their maturity. I don't think there are any bond indexes to track going back far enough. Maybe one can be simulated (I guess the ladder does this?)

Or did I miss something?

-ERD50

Rebalancing?

I don't see how you can accurately rebalance each year unless your bond allocation reflects capital appreciation/loss.

 

[CaliforniaMan]

I'm not so sure this is a problem. But it raises a question - What should FIRECalc assume?
...
It seems complicated, it isn't just the NAV of 5 year treasuries, it has to do with their maturity. I don't think there are any bond indexes to track going back far enough. Maybe one can be simulated (I guess the ladder does this?)

Or did I miss something?

-ERD50

Rebalancing?

I don't see how you can accurately rebalance each year unless your bond allocation reflects capital appreciation/loss.

I observed a similar big difference in a test case, as reported in the other thread. See below.
...

FIRECalc offers the choice of 30-yr and 5-yr Treasury. However, the problem is that bonds are not like stocks. With the latter, all shares are fungible, at least the ones of the same class. Not so with bonds. Individual bonds have different coupon rates, different maturity dates, so how do we model them all?
...
So, modeling individual bonds is going to be tough, if we want to mark them to market value. With the MF, perhaps there are existing bond indices that can be used. However, these indices may not go back enough to study historical portfolio performance.

I guess FIRECalc uses the published historical coupon rates to compute the bond income. This however does not match the income of either the individual bond buyer, nor the MF share holder. And it misses the market value of the principal, which is important for a portfolio balancer.

What to do?

For the question of what should FIRECalc do, I would vote for including duration. As I understand it (and I am not an expert) any portfolio of bonds has a duration, which is a function among other things, of the bond's interest rate and its maturity. It measures the number of years it takes to recover from an interest rate change. So for example for a 5 year duration, a 1% change in interest rates will change the value of the bond by 5%.

If we own a portfolio of bonds, we can calculate the portfolio duration. If we own bond funds, we can use the durations given by the fund.

It seems to me that this is essential for re-balancing calculations to be modeled correctly. FIRECalc could offer an option to include your duration and construct the bond principal changes from that.

However there is one other question and that is what interest rate to use? The interest rate used in calculating the duration will differ from the 30 year or 5 year treasuries included in FIRECalc, and of course the rate curve varies over time.

But even if we can't have a perfect interest rate, IMO an estimate, using either the 30 year, 5 year or some function of them, would be better than what FIRECalc is doing right now, ignoring principal changes.

I wonder how the other calculators are handling this, if at all.

 

[ERD50]

Rebalancing?

I don't see how you can accurately rebalance each year unless your bond allocation reflects capital appreciation/loss.

True. I was going to add that rebalancing doesn't seem to have a big effect on actual returns/survive-ability according to some studies. It might make people feel good by maintaining an AA they are comfortable with, but the selling on the way up seems to negate any real benefit. But did those studies take into effect the bond NAV changes?

While it really should not be ignored, I'm guessing that it probably pretty much washes out over time. While NW-B's example shows a very significant delta in the 2001-2011 time frame with a 100% bond portfolio, will this make a big difference over the long term, with a more typical 25-50% bond allocation?

... Like ERD I am willing to assume that I am not unlucky enough to be faced with inflation, market performance etc that are worse than the last hundred plus year. ...

To clarify this, I really do not make that assumption. I have no reason to believe that the future can't be worse than the worst of the past. So I shoot for a 100% historically safe WR and throw a little buffer on that, and will just have to make adjustments if things go even beyond that. If you think in terms of 100+ data points, those worst cases can be considered outliers. But if you think in terms of those failures representing a handful of business cycles in those 100+ years, then maybe those are pretty common?

What I have said several times is that even though I agree that the average returns going forward may well be below past averages, that does not mean that the future will be worse than the worst of the past. And FIRECalc focuses you on the worst of the past, averages really don't enter into it.

So I find Monte Carlo simulators to be really dangerous forecasting tool, perhaps especially in the hand of experienced operators. ...

Agreed. Who is to say what are the 'right' terms to use? Using the weather analogy, would you trust a model you built to predict future high temperatures for a city? Would you compare it to historical data to see if it makes sense?

-ERD50

 

[Lsbcal]

...(snip)... When it comes to corporate or government bonds, I buy an MF. A 30-yr bond MF like VUSTX shows a duration of around 15 years. I guess it has a ladder of bonds, hence the average duration is 1/2 of that of the newest bond in its portfolio. The yield is the average of all the individual bonds in the portfolio.

So, modeling individual bonds is going to be tough, if we want to mark them to market value. With the MF, perhaps there are existing bond indices that can be used. However, these indices may not go back enough to study historical portfolio performance.
...

I too buy bond funds mostly for the FI part of my portfolio. Mostly I hold intermediate bond funds with a duration of 4 to 5 years. Hence my using the 5 year Treasury choice in FIRECalc.

For my own calculator I've used the constant maturity series from the Fed as shown here: 5-Year Treasury Constant Maturity Rate (GS5) - FRED - St. Louis Fed
This monthly series goes back to only 1953. I can then calculate the monthly interest plus rate change component by just using:
return = (1 + interest rate)^(1/12) + duration*(monthly rate change)

I'm really unhappy with FIRECalc leaving out the second term and others above have expressed. Most investors count on the smoothing effect of bonds going up when equities decline and rates move down. All the focus on Wellington and Wellesley funds have to do with their nice balancing act which ties in with that rate sensitivity (for good and bad).

 

[NW-Bound]

...
While it really should not be ignored, I'm guessing that it probably pretty much washes out over time. While NW-B's example shows a very significant delta in the 2001-2011 time frame with a 100% bond portfolio, will this make a big difference over the long term, with a more typical 25-50% bond allocation? ...

The example I showed was for a 50/50 blend of Total Stock/Long Treasury. See post #9 above, which was taken off the other thread.

To repeat, that example shows a portfolio end balance of $1183K using two Vanguard MFs, which is a 66% increase from the $714K as computed by FIRECalc.

 

[ERD50]

The example I showed was for a 50/50 blend of Total Stock/Long Treasury. See post #9 above, which was taken off the other thread.

To repeat, that example shows a portfolio end balance of $1183K using two Vanguard MFs, which is a 66% increase from the $714K as computed by FIRECalc.

Sorry, my mistake. I was actually referring to post #2 by Lsbcal.

...

So I did a test for years 2002 to 2011. I looked at a pure 5 year Treasury portfolio of $1M with no spending. FIRECalc gave a final value of $1,064,095 inflation adjusted. Then I looked at what $1M would have done in VFIUX (Vanguard 5 year Treasury admiral). The final inflation adjusted value was $1,457,944 (for total return, green area below). This is a huge difference. Then I just used the income return for VFIUX (blue area below). The result was much closer to FIRECalc $1,173,121 indicating FIRECalc does indeed ignore the capital returns. ....

-ERD50

 

[Lsbcal]

Hi ERD50, we talk about sequence of returns being important in ER and to ignore this smoothing component when equities fall out of bed is not at all something I would want to do.

I do have my own calculator which is not general enough to suggest to others. It uses the income + capital return components and computes on a monthly basis.

 

[photoguy]

What did the original Bengen / trinity studies do for the bond portion?

+1 to clifp's comments on MC. Whenever I read an MC swr paper I always think it tells me more about the author (and his/her outlook) than anything useful about the future.

 

[Lsbcal]

I hope I'm not beating this thing to death but I wanted to see just how far off FIRECalc might be in a typical balanced portfolio case. So I compared a simple 65/35 portfolio of total stock market and 5 year Treasuries. I'm not sure what FIRECalc uses for the 5 year Treasuries but if I were building a portfolio I'd probably use Vanguards Intermediate Treasuries.

Here are the results for 65/35 AA and 3.3% spending
(1) Blue area: Total Stk Mkt Admiral (VTSAX) and Intermediate Treasury Fund (VFITX) with only the income component
(2) Orange area: Same as (1) but using VFITX's total return
(3) Tan area: FIRECalc with it's total stock market and 5 year Treasury

Clearly FIRECalc (tan area) results are a lot different then the mutual fund results (orange area). Of course, I could have botched the numbers so this would need some checking.

 

[clifp]

No I don't think you are beating it do death all. I think it is important to bound how big an error we are looking at. If I am reading your chart correctly it says Firecalc is showing a portfolio of $530K (I assume in 2000 dollars) where as the correct amount using actual mutual funds is $810K

 

[REWahoo]

What did the original Bengen / trinity studies do for the bond portion?

I'm really interested to know the answer to this question as well.

 

[ERD50]

I hope I'm not beating this thing to death ...

No I don't think you are beating it do death all. I think it is important to bound how big an error we are looking at. ...

Agreed. Even though I was thinking this might wash over the long run, I could be wrong about that, and we really should understand it. Those are some big deltas.

-ERD50

 

[Lsbcal]

No I don't think you are beating it do death all. I think it is important to bound how big an error we are looking at. If I am reading your chart correctly it says Firecalc is showing a portfolio of $530K (I assume in 2000 dollars) where as the correct amount using actual mutual funds is $810K

That's right and also the minimums are shown in red for 2008. I always like to look at how bad things could have been.

 

[ERhoosier]

What did the original Bengen / trinity studies do for the bond portion?

I'm assuming this was the "original" Bengen study (J Fin Planning, Oct 94).
http://www.retailinvestor.org/pdf/Bengen1.pdf

The article says "intermediate term Treasuries" (10yr ?) were used for bond portion of portfolio, and it appears that changes in bond prices were NOT included in the analysis. In the article's "appendix" (p 179) Bergen states he assumes the portfolio is in tax-deferred account (i.e. no tax effect) & describes his method of rebalancing. His example- $1M porfolio with AA of 50/50 stocks/bonds. Yr 1 returns stocks 10%, bonds 5%, and EOY withdrawal (4%SWR @ 3% inflation).
EOY Yr 1= $550k stocks + $525k bonds - $41,200 withdrawal= $1,033,800 EOY balance, or starting Yr 2 with $516,900 in both stocks/bonds. In his Yr 1 rebalance example, Bergen mentions nothing about changes in bond prices nor trading bonds so I believe his general assumption is holding to maturity with no default/credit risk (Treasuries).

I don't know of any complex analysis of Bergen's theory which includes realized capital gains/losses in bonds, but IMHO it would likely irrelevant to his analysis. Since annual returns of stocks are generally higher than bonds, rebalancing in most years would mean net purchases of bonds (assuming rolling maturation of laddered bond portfolio). Significant changes in bond prices would be relevant only when bonds (int term Treasuries) needed to be sold to bump up stock AA. This would typically occur after a year of markedly declining stock prices, a scenario during which bond prices usu rise (falling int rates in weaker economy). But there have been exceptions to this (e.g. 69-70, 73-5) when bonds might have been sold at loss. Conversely, since rates have been generally falling since early '80's, bonds sold for rebalancing since then would likely have been sold at gains. My guess is that in Bergen's long term SWR portfolio any net bond price gains/losses would be small.

 

[clifp]

I'm assuming this was the "original" Bengen study (J Fin Planning, Oct 94).
http://www.retailinvestor.org/pdf/Bengen1.pdf

The article says "intermediate term Treasuries" (10yr ?) were used for bond portion of portfolio, and it appears that changes in bond prices were NOT included in the analysis. In the article's "appendix" (p 179) Bergen states he assumes the portfolio is in tax-deferred account (i.e. no tax effect) & describes his method of rebalancing. His example- $1M porfolio with AA of 50/50 stocks/bonds. Yr 1 returns stocks 10%, bonds 5%, and EOY withdrawal (4%SWR @ 3% inflation).
EOY Yr 1= $550k stocks + $525k bonds - $41,200 withdrawal= $1,033,800 EOY balance, or starting Yr 2 with $516,900 in both stocks/bonds. In his Yr 1 rebalance example, Bergen mentions nothing about changes in bond prices nor trading bonds so I believe his general assumption is holding to maturity with no default/credit risk (Treasuries).

I don't know of any complex analysis of Bergen's theory which includes realized capital gains/losses in bonds, but IMHO it would likely irrelevant to his analysis. Since annual returns of stocks are generally higher than bonds, rebalancing in most years would mean net purchases of bonds (assuming rolling maturation of laddered bond portfolio). Significant changes in bond prices would be relevant only when bonds (int term Treasuries) needed to be sold to bump up stock AA. This would typically occur after a year of markedly declining stock prices, a scenario during which bond prices usu rise (falling int rates in weaker economy). But there have been exceptions to this (e.g. 69-70, 73-5) when bonds might have been sold at loss. Conversely, since rates have been generally falling since early '80's, bonds sold for rebalancing since then would likely have been sold at gains. My guess is that in Bergen's long term SWR portfolio any net bond price gains/losses would be small.

While I agree over a 30 year period the capital gains/losses for bonds would modest. The thing is what is important is not the average but the worse case, when we are looking at survival rates.

We already know that 1929 was a bad time to retire cause the stock market crashed, but if you had a significant bond portfolio with 10 and/or 30 year treasury they increased in value making rebalancing less painful. Interest rates in the early 2000 were flat to modestly declining once making it easier to survive with a 4% withdrawal. The same think is true in 2008 10 year Treasury started of at 3.74% and hit a low of 1.91% in 2012. If you rebalanced Jan 1 of 2014 you sold stocks and bought bonds at a 2.97%. Or you got 4% more Total bonds shares Jan 2014 than in Jan 2013.

If we look at the methodology of FIRECalc, Raddr, Trinity studies, and suspect most Monte Carlo, prospect for a Y2K retire look really grim. But if you look at what NW and Lsbcal did not nearly so bad.

Perhaps just as importantly if you look at my portfolio, or another 2000 retiree like MichealB (I am guessing about his portfolio) the situation is considerably better.

I sort of hate to see people anxious to retire, hit their number and then convince themselves that 4% isn't safe when it maybe be even more safe than we realize cause the calculators have flaws.

Conversely, I think it is, more likely than not, that sometime in the next 5 years we have a bear market, accompanied and perhaps even caused by higher interest rates. If the high interest rates also include high inflation than 201x may actually be one of the times when a 4% withdrawal isn't safe. But the higher interest rates may mask the decline in the bond portfolio and create a false sense of security for retirees of the 2020s looking back.

I contributed in the past to FIRECalc (although can't seem to login) and I'd be willing to do so now if we can make it a more accurate and useful tool.