How to model TIPS rates (and returns) pre-1997

Discuss all general (i.e. non-personal) investing questions and issues, investing news, and theory.
longinvest
Posts: 5672
Joined: Sat Aug 11, 2012 8:44 am

Re: How to model TIPS rates (and returns) pre-1997

Post by longinvest »

I have uploaded a new version (0.2) of the TIPS simulator spreadsheet. The links are in the following post:
viewtopic.php?f=10&t=203092&p=3138674#p3132364

Here's a summary of the changes:
  • Use December BLS CPI numbers.
  • Use 2-years trailing inflation as expectation.
  • Use a weighted average: (2 X (fund 30-2) + 1.5 X (fund 10-2) + (fund 5-2)) / 4.5
Here are the justifications for the changes.

CPI Data

Siamond noticed that I was using January BLS CPI numbers. I had initially used these numbers as this is what Shiller had extracted off the BLS data starting in year 1913. But, as we calculate end-of-December to end-of-December returns everywhere, it makes more sense to use the December numbers.

Inflation Expectation

A quick comparison of the 2003-2015 sythetic returns with real-world returns reveals that using trailing 2-year inflation is a better match than 3 years or 1 year. I know, this is curve fitting. Complain if you want. :annoyed

Weighted Average

OK, you'll accuse me of curve fitting, again, but this time I will defend my position. I had initially used a simple average of the (30-2), (10-2), and (5-2) fund returns to estimate total-market returns, based on the idea that the Treasury issues 30-year, 10-year, and 5-year TIPS.

But, the problem is that the (30-2) fund already includes approximately 30% of 10-year or less TIPS, making the average weightings of 10-year and 5-year TIPS pretty high in the computed average. So, I tried, instead, to add 2 weights of (30-2), 1.5 weight of (10-2) and 1 weight of (5-2). Here's how the results compares with real-world returns (using December CPI and FRED TIPS yields):

Image

You'll agree that this is a better match (given exact FRED yields) to real world returns than our previous (equal-weighted, using January CPI) attempt:

viewtopic.php?f=10&t=203092&p=3138674#p3132364
longinvest wrote:Image
OK, OK. I see the burning question on your lips: "How does the updated synthetic TIPS model compare to real-world returns?"

Here's the answer:

(trailing 2-year inflation as expectation, December CPI, weighted average)
Image

The question is: Are my changes legitimate?

What do you think?
Last edited by longinvest on Thu Apr 12, 2018 7:01 am, edited 1 time in total.
Variable Percentage Withdrawal (bogleheads.org/wiki/VPW) | One-Fund Portfolio (bogleheads.org/forum/viewtopic.php?t=287967)
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

I need to run, a few quick points:

* I don't think data mining is such a big sin here, as we're trying to recreate past history, and not create a model helping us with current or future decisions.

* better approximating the way a total-TIPS fund work(ed) is of course an excellent idea. Longinvest, let me suggest you take a look at Figure 1 or the "the long and short of TIPS" article from Vanguard, and this will give you some historical perspective.

* seems to me that we need a separate model to estimate past BEI (break-even inflation) rate more than inflation, as an input to this great work from longinvest. This is close to 'expected inflation' of course, but the point is that we should aim at matching the BEI actuals (from the past, 1998 till now), as they were themselves not necessarily matching very well with realized inflation (otherwise TIPS returns would be roughly equal to TBM returns, and well, they were not). I'll work more on this.

* would be lovely to be able to test our findings out of sample with UK numbers... Third step!
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

I gave it a quick try with the latest model from longinvest. I compared:
P1. as a reference, the Barclays US TIPS index
P2. the Vanguard fund, VIPSX
P3. the model with expected inflation == past 2 years (default choice)
P4. the model with expected inflation == past year (was my proposal to start with)
P5. the University of Michigan consumer survey (1yr and 5yr expected inflation, starting in 1966; spliced with P2 for earlier years)

Here are the usual charts, with the telltale showing P3 and P4, relative to the Barclays index. Note that I didn't show VIPSX as its trajectory is nearly indiscernible from its index (very low tracking error of 0.32, how is that for an active fund!). As you can see, both models (P3 and P4) do ok, albeit with rather strong variations up and down. It is also interesting to see how different their respective trajectories are, those TIPS are VERY sensitive to their initial yield (i.e. their pricing), which is a direct function of the expected inflation. I would tend to agree with longinvest's choice of P3 as a decent default, as the std-deviation for P4 is just too high.

Image

Then I added the UMich model to the telltale (and removed model P4 for readability). Well, it's actually fairly decent, except for the first couple of years. It seems to me that we're definitely onto something here, but really need to refine a tad the inputs, namely the 'expected inflation'.

Image

PS. as to the papers from Prof. Kothari and Groen, they unfortunately do not provide results for those years where TIPS existed for real. So we can't compare, which is too bad.
longinvest
Posts: 5672
Joined: Sat Aug 11, 2012 8:44 am

Re: How to model TIPS rates (and returns) pre-1997

Post by longinvest »

Siamond,

Thanks for the analysis.

Here's one thing I have been thinking about. It is to simply do the reverse calculation on FRED TIPS yields. In other words, extract zero-coupon TIPS yields out of them (like I did with nominal yields) and then recover the inflation expectation across maturities to be able to see the shape of the expectation curve. Is it constant, increasing, decreasing, something else? How does it evolve over time?

The crude "2-years trailing inflation" expectation is used as a constant across all maturities. Maybe we could improve on this just by shaping the curve closer to reality?

I did the calculations. Here's what I get:

Inflation Expectations (zero nominal coupon vs zero real coupon)
Image

As you can see, the expectations are not constant across maturities. (Remember that the FRED TIPS yields other than 30, 20, 10, and 5 were linearly approximated. For nominal yields, 30, 20, 10, 7, 5, 3, 2, and 1 yields were available, the rest were linearly approximated).

Is there any pattern, in there?
Last edited by longinvest on Thu Apr 12, 2018 7:00 am, edited 1 time in total.
Variable Percentage Withdrawal (bogleheads.org/wiki/VPW) | One-Fund Portfolio (bogleheads.org/forum/viewtopic.php?t=287967)
stlutz
Posts: 5585
Joined: Fri Jan 02, 2009 12:08 am

Re: How to model TIPS rates (and returns) pre-1997

Post by stlutz »

Some thoughts after trying to waterboard this data some and not getting anywhere too satisfactory.

--I think trying to overfit the last 15 years only takes one so far. Actual inflation hasn't been all that volatile and if you look at a chart of the breakeven rate, it really doesn't change all that much if you block out that big drop in '08 (https://fred.stlouisfed.org/graph/?g=c3Sp). Trying to model that '08 period is problematic because it's generally understood that the big drop in TIPS prices that year was due in part to declining inflation expectations, but the extreme volatility was driven by liquidity issues in the TIPS market. Trying to come up with an expected inflation model to replicate a liquidity-driven event probably doesn't make sense. Perhaps trying to fit the years not including '08 and '09 makes sense.

--I spent some time looking at the 1970s and early 80s where inflation was moving around quite a bit. And the returns we end up with there by using the past couple years of inflation are a problem because you are getting huge changes in expected forward inflation year to year which produces results like a 63% return in 1974. I'm noticing that the modeled returns for these two decades have a much higher standard deviation than the ones from Prof. Kothari.

--Conceptually, the nominal yield curve should help with estimating a Treasury-market-expected inflation series. Future inflation expectations are one factor (among several) in how the yield curve looks. A steep yield curve indicates that the market expects higher inflation in the future; a flat or inverted curve indicates the market expects declining inflation. So, for example, if I compare the 2 vs. the 20 year nominal, in the late 70s and early 80s, the yield difference was very small, which indicates a market belief that inflation was heading lower; in 2010-2013, the curve was very steep, indicating higher expected inflation in the future. The problem with this is that the are multiple factors that go into nominal yields, so it's really not possible to extract out an inflation expectation. As a result, I haven't been able to turn this into a calculation that works any better.

One of the reasons for the existence of TIPS is to find out a market expectation of inflation. So, we are trying to solve for something that the experts thought couldn't be determined without creating a new type of security.

Thanks for your efforts. Sorry I wasn't able to provide more fruitful assistance.
dcabler
Posts: 4482
Joined: Wed Feb 19, 2014 10:30 am

Re: How to model TIPS rates (and returns) pre-1997

Post by dcabler »

There's another paper out there that attempts to create a synthetic tips sequence

This is the original, though I had some issues downloading it today.
https://www.bostonfed.org/-/media/Docum ... er199a.pdf

It appears to be available here as well: http://www.docstubo.net/41693814/Neer199a-pdf/
dcabler
Posts: 4482
Joined: Wed Feb 19, 2014 10:30 am

Re: How to model TIPS rates (and returns) pre-1997

Post by dcabler »

dcabler wrote:There's another paper out there that attempts to create a synthetic tips sequence

This is the original, though I had some issues downloading it today.
https://www.bostonfed.org/-/media/Docum ... er199a.pdf

It appears to be available here as well: http://www.docstubo.net/41693814/Neer199a-pdf/
There's also the Ibbotson/Morningstar paper: http://philgmh.tripod.com/TIPS.pdf
Interesting - it uses the standard deviation of the previous 36 months of inflation as a term in the calculation....
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

A bit of an update here. Sadly not in a good way.

1. Longinvest did really cool work with his spreadsheet, but this is only truly useful if we find a proper data series for historical expected inflation (aka BEI) that would have reasonably matched the break-even rate of TIPS, if they had existed pre-1997. Otherwise it's kind of garbage in, garbage out.

2. The 'last 1 or 2 or 3 years of inflation == expected inflation' basic data series is just that, very basic. Plus let me quote longinvest:
"[..] bad idea [..] it would transform our synthetic TIPS into clunky nominal bonds, by simply dephasing inflation factors between nominals and TIPS."

3. It is also difficult to meaningfully compare against actuals since inflation was so tame in the past couple of decades (plus we have this weird 2008 hiccup about TIPS returns, which appears due to 'other reasons', e.g. liquidity, and maybe the issue is broader than 2008). As stlutz said:
"trying to overfit the last 15 years only takes one so far."

4. Re-reading the work from Prof. Kothari/Shanken as well as the work from Groen/Middeldorp, it seems to me that the overarching logic for their expected inflation model does a LOT of ex post analysis (regression analysis on multiple factors, etc). It is hard to believe that this would align with the way a TIPS purchaser (or the market as a whole) might assess expected inflation ex ante.

5. I played a bit with the UMich data, but it is poorly documented, it appears inconsistent, a year is missing, this doesn't inspire much confidence. Plus polling techniques as a way to assess expected inflation? Really? Maybe we had enough issues with polls reliability recently...

6. Back to Kothari and Groen, it is hard to get a pragmatic sense of how valid their models are, as they didn't provide spreadsheet data for years of known TIPS returns. We know that the Kothari detailed model was lost in time. I issued a request to Groen/Middeldorp, asking if we could get access to their detailed model so that we could extend it towards known years (and asking a couple more questions), and... no luck.

7. I also couldn't find any further information about the Bridgewater study. I didn't pursue the Ibbotson/Chen track, or some other pointers. I'll let other people pursue those avenues if they feel like it.

8. Oh, and longinvest seems to have some qualms about the logic of his own spreadsheet (just when I was starting to like it!). I'll let him elaborate if he feels like it.

So... bottomline is that we at least succeeded to get the actual numerical findings of Kothari/Shanken (instead of eye-balling a chart), complemented by some Barclays index numbers we didn't use before, that's progress, but... we may not be able to go much further than that.

I did copy those updated numbers in the working version of the Simba spreadsheet. And I will retain the 'Synthetic TIPS' name for the data series, as it nicely captures that said data series should really be taken with a giant grain of salt.
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

Oh, and I also meant to show a comparison between the primary models we have so far for the pre-VIPSX years (i.e. 1970 to 2000). The reference is 'cash' (0% return), which means that the telltale chart is actually a growth chart. A small trick I recently added to the Simba spreadsheet. :wink:

Similar set of portfolios as I presented before, P2 to P5 with the same ER (0.20 like VIPSX today):
P1. as a reference, cash
P2. the current pre-VIPSX data series (i.e. Kothari until 1996, Barclays TIPS index until 2000)
P3. longinvest's model with expected inflation == past 2 years (default choice)
P4. longinvest's model with expected inflation == past year (my old -bad- proposal)
P5. longinvest's with the University of Michigan consumer survey (1yr and 5yr expected inflation, starting in 1966; spliced with P2 for earlier years)

I removed P4 from the telltale/growth chart for readability. Note how the CAGR can change by one full point, depending on the expected inflation inputs. Also it is interesting (and maybe a good sign? or just plain luck?) that the Kothari/VIPSX series is reasonably well aligned with the UMich series.

Image
longinvest
Posts: 5672
Joined: Sat Aug 11, 2012 8:44 am

Re: How to model TIPS rates (and returns) pre-1997

Post by longinvest »

siamond wrote: 8. Oh, and longinvest seems to have some qualms about the logic of his own spreadsheet (just when I was starting to like it!). I'll let him elaborate if he feels like it.
There are two parts, to the Synthetic TIPS spreadsheet.

One part for which the logic works really well. That part is the TIPS fund simulation part. When we feed it historical real yields extracted from FRED (along with a linear approximation of missing yields), we get approximate TIPS fund returns that are pretty similar to real-world fund return:

viewtopic.php?p=3159772#p3139060
longinvest wrote: Image
That's pretty good!

The second part for which the logic is broken. It's the part which attempts to extract real yields from nominal yields using different approximations of inflation expectations. Logically, if there's any influence of one type of yields on the other, in real life, it should be the other way around: nominal yields being based on real yields and expected inflation. If we had a good approximation of break-even inflation, there wouldn't be any problem. But, it appears that the relation between nominal yields and real yields is not as simple as one would think.

As Stlutz wrote earlier:

viewtopic.php?p=3159772#p3139980
stlutz wrote:One of the reasons for the existence of TIPS is to find out a market expectation of inflation. So, we are trying to solve for something that the experts thought couldn't be determined without creating a new type of security.
All of our attempts, in post 1997 years, at approximating expected inflation failed miserably to give anything near real-world returns. Here's a typical outcome we got, despite inflation being tame and pretty stable in the last two decades:

viewtopic.php?p=3159772#p3139060
longinvest wrote: Image
(Don't be fooled by the fact that the curves join back together in the last part of the chart, it's a fluke. Different assumptions lead to different results.)

Actually, in a later post, I extracted linearly approximated real-life inflation expectations across the yield curve, based on FRED nominal and TIPS yields, and I got the following chart where I am unable to detect any stable pattern that could be used to guess the shape of historical inflation expectations across the yield curve:

viewtopic.php?p=3159772#p3139955
longinvest wrote: Inflation Expectations (zero nominal coupon vs zero real coupon)
Image

As you can see, the expectations are not constant across maturities. ...
In other words, not only we don't have a good estimate of inflation expectations, worse, we would have to guess the shape of these expectations across the yield curve to be able to reconstruct trustworthy TIPS fund returns. It's just too much speculation about what could have happened with TIPS before their introduction into the market, let alone that we're assuming that their existence wouldn't have disturbed the returns of other assets in the market (in other words: zero market impact). That's just too much for me.
siamond wrote:I did copy those updated numbers in the working version of the Simba spreadsheet. And I will retain the 'Synthetic TIPS' name for the data series, as it nicely captures that said data series should really be taken with a giant grain of salt.
Yes, a humongous grain of salt. So huge that I would personally vote for the removal of synthetic TIPS from the Simba spreadsheet, if there was a vote on it. I think that including the synthetic series does not serve any useful purpose, unless there's a desire to mislead users. People using the spreadsheet expect it to be relatively reliable; I can't say that I have any confidence whatsoever in the adequacy of synthetic TIPS returns to represent what could have happened in pre-TIPS years (and in TIPS years).

For one thing, I don't think that the Kothari/Shanken work modeled a fund with the kind of precision that we have (modeling different rungs). For another, I don't have confidence that they were able to find good inflation expectations that would have worked past 1997, otherwise they would have been proud to include them in their paper. And, lastly, I am sure that they didn't find the pattern of inflation expectation across the yield curve. That would have been a major accomplishment that they would have documented, probably in a paper of its own.

If we could, at least, have validated that the chosen synthetic model works well enough to guess TIPS returns after 1997, we could have some minimal level of confidence in the adequacy of the synthetic returns. Without this, I think that we're just leaving random data in the spreadsheet that will just mislead users.

I won't insist on it. I was just expressing my personal opinion.

Anyway, I'd like to thank Siamond for all his help and comments on the (unfortunatly) failed project of modeling synthetic TIPS.

Siamond is doing all Bogleheads a huge service with all his work on cleaning up Simba's data. Kudos!
Last edited by longinvest on Thu Apr 12, 2018 6:59 am, edited 1 time in total.
Variable Percentage Withdrawal (bogleheads.org/wiki/VPW) | One-Fund Portfolio (bogleheads.org/forum/viewtopic.php?t=287967)
User avatar
Kevin M
Posts: 15750
Joined: Mon Jun 29, 2009 3:24 pm
Contact:

Re: How to model TIPS rates (and returns) pre-1997

Post by Kevin M »

longinvest wrote: Yes, a humongous grain of salt. So huge that I would personally vote for the removal of synthetic TIPS from the Simba spreadsheet, if there was a vote on it. I think that including the synthetic series does not serve any useful purpose, unless there's a desire to mislead users.
Agreed.

However, it would be nice to archive efforts like this in a way that future Bogleheads who might take up the mantle of maintaining and improving the backtest spreadsheet could learn from it and avoid going down blind alleys.
Anyway, I'd like to thank Siamond for all his help and comments on the (unfortunatly) failed project of modeling synthetic TIPS.

Siamond is doing all Bogleheads a huge service with all his work on cleaning up Simba's data. Kudos!
Also agreed, and also kudos to you for your contributions, longinvest.

Kevin
If I make a calculation error, #Cruncher probably will let me know.
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

longinvest wrote:Yes, a humongous grain of salt. So huge that I would personally vote for the removal of synthetic TIPS from the Simba spreadsheet, if there was a vote on it. I think that including the synthetic series does not serve any useful purpose, unless there's a desire to mislead users.
I do not disagree at all... :wink:

Yes, I actually reached a similar conclusion about Emerging Markets after quite some meandering (will explain in a more proper thread). And I agree, the same holds true here, models are just too unreliable. I wouldn't want to give up on tracking TIPS though. I believe the Barclays index is a solid input, so we have solid numbers since 1997. We could restrict the use of the Kothari series to 1985+ and move TIPS to the set of data series starting in 1985, with the corresponding charts and metrics. This would make for 12 years of dubious data, and ~20 years of solid data. Which is far from satisfying, admittedly.

I would observe though that the time of wild inflation was done by 1985, and the more steady, Fed-controlled, time started more or less at that point.

Hm, I'm torn here. I agree we should can the 1970-1984 data (I can leave a pointer to the full Kothari spreadsheet just for reference). I am more hesitant about restricting the TIPS series to a small corner (1997+) where nobody will look at it anymore...

PS. I may have another approach, actually, kind of blunt... Let me check a few numbers...
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

siamond wrote:PS. I may have another approach, actually, kind of blunt... Let me check a few numbers...
Oh boy. I was playing with the idea of bluntly approximating TIPS returns as IT bonds returns for the 1985-1996 period. I changed my mind. This is... ridiculous. I'll let you appreciate what I mean. The indices are the benchmarks followed by the corresponding Vanguard funds (which are active). Tracking error is relative to the VIPSX data series (it's roughly an RMSE formula). Index series do not include any ER adjustment. Blue lines express my puzzlement.

Image

(edit: added the 10-yrs treasury rate, just out of curiosity; added ST treasuries; made VIPSX the reference)
Last edited by siamond on Wed Dec 21, 2016 11:13 pm, edited 3 times in total.
stlutz
Posts: 5585
Joined: Fri Jan 02, 2009 12:08 am

Re: How to model TIPS rates (and returns) pre-1997

Post by stlutz »

My opinion remains to use the *short-term* Treasury series prior to 1997. Prior to then, if you wanted to guard your bond portfolio against inflation, this is what you used--that was the only actual option available at the time. With less duration risk, you were less exposed to increases in inflation.

It is problematic to combine two different types of data series together into one, but the longer series is a way to assess whether there was value in focusing on the impact of inflation to the bond portion of a balanced portfolio.

And to take a bigger picture look, I think very recent TIPS returns illustrate what happens when you have the combination of inflation being lower than expected and inflation expectations going lower (i.e. nominals having good returns and TIPS having negative returns). At least some years in the 70s were cases where the reverse was happening. The short-term Treasury series does provide better returns than intermediate term bonds or Total Bond in those rising-inflation years, meaning that you do at least have a balanced look across different regimes.

Far from ideal, but I think it's better than including any [very intelligently] made-up data series.
User avatar
Tyler9000
Posts: 739
Joined: Fri Aug 21, 2015 11:57 am

Re: How to model TIPS rates (and returns) pre-1997

Post by Tyler9000 »

longinvest wrote:
siamond wrote:I did copy those updated numbers in the working version of the Simba spreadsheet. And I will retain the 'Synthetic TIPS' name for the data series, as it nicely captures that said data series should really be taken with a giant grain of salt.
Yes, a humongous grain of salt. So huge that I would personally vote for the removal of synthetic TIPS from the Simba spreadsheet, if there was a vote on it. I

For one thing, I don't think that the Kothari/Shanken work modeled a fund with the kind of precision that we have (modeling different rungs). For another, I don't have confidence that they were able to find good inflation expectations that would have worked past 1997, otherwise they would have been proud to include them in their paper. And, lastly, I am sure that they didn't find the pattern of inflation expectation across the yield curve. That would have been a major accomplishment that they would have documented, probably in a paper of its own.

If we could, at least, have validated that the chosen synthetic model works well enough to guess TIPS returns after 1997, we could have some minimal level of confidence in the adequacy of the synthetic returns. Without this, I think that we're just leaving random data in the spreadsheet that will just mislead users.
First of all, thanks to Siamond, Longinvest, and the rest of the crew for doing such amazing work on this. It's very impressive and a true service to the investing community.

Synthetic TIPS definitely look like a mess. It's kinda weird to think about needing to predict the future to reconstruct known history, but TIPS are a funny animal. And lacking the proper crystal ball any effort to model them is going to be difficult if not impossible. While I'm no expert on TIPS modeling methodology and I'll defer to the wisdom of others on that, I do find Longinvest's issues with the Kothari paper compelling. Enough so to make me question whether we can confidently assume it truly contains the data we're looking for.

As for just using another bond type prior to 1997, I'm working on some tools to directly compare two series for how well they track one another over time. It would take too long to explain in a post like this, but the short story is that I agree that intermediate treasuries are a poor fit as a direct stand-in. Short term treasuries are really no better. I'll keep experimenting, but I'm not confident that will bear fruit with a reasonable amount of error to represent as the same thing. We are all free to substitute similar but unrelated assets on our own, but I wouldn't vote for doing it by default.

Personally, I value the Simba spreadsheet for its integrity. I hate removing data, but I like knowing that people have studied the data inside and out in threads like these and feel good about it. That type of honesty and peer review is exactly what makes it so useful. So no matter the final decision, I appreciate all the work that has gone into this.
dcabler
Posts: 4482
Joined: Wed Feb 19, 2014 10:30 am

Re: How to model TIPS rates (and returns) pre-1997

Post by dcabler »

Thanks longinvest and siamond for all of the heavy lifting for this work! I agree that it's a mess and truly a difficult task. Reading through as much as I've been able to digest, I can appreciate the difficulty of creating a model for this as there does appear to be some sort of mutual dependence between regular treasuries and Tips.

Without arguing for the validity of having TIPs in either an accumulation or decumulation portfolio, my vote would be to keep a series in the spreadsheet along with the grain/block of salt disclaimer about the accuracy in the pre-1997 years.

siamond - I know at one time you were trying to get data for tips-equivalents from other countries in order to get more insight into how they have historically operated elsewhere. Any success there?

Thanks again all!
dcabler
Posts: 4482
Joined: Wed Feb 19, 2014 10:30 am

Re: How to model TIPS rates (and returns) pre-1997

Post by dcabler »

Vanguard has an interesting paper on inflation expectations. Interesting results on hedging expected and unexpected inflation. In the appendix is their methodology of how they break down the two with a linear regression plus out-of-sample testing...

https://personal.vanguard.com/pdf/icruih.pdf
longinvest
Posts: 5672
Joined: Sat Aug 11, 2012 8:44 am

Re: How to model TIPS rates (and returns) pre-1997

Post by longinvest »

Dcabler,
dcabler wrote:Reading through as much as I've been able to digest, I can appreciate the difficulty of creating a model for this as there does appear to be some sort of mutual dependence between regular treasuries and Tips.
No, there does not appear to be an obvious mutual dependence between nominal and real yields (Treasuries and TIPS).

We initially thought that there was a simple one, but reality showed us otherwise. The relation between nominal yields and real yields does not appear to be remotely stable across time, and, to compound the problem, it varies across the yield curve at any point in time with no discernible pattern.
Variable Percentage Withdrawal (bogleheads.org/wiki/VPW) | One-Fund Portfolio (bogleheads.org/forum/viewtopic.php?t=287967)
longinvest
Posts: 5672
Joined: Sat Aug 11, 2012 8:44 am

Re: How to model TIPS rates (and returns) pre-1997

Post by longinvest »

siamond wrote:
siamond wrote:PS. I may have another approach, actually, kind of blunt... Let me check a few numbers...
Oh boy. I was playing with the idea of bluntly approximating TIPS returns as IT bonds returns for the 1985-1996 period. I changed my mind. This is... ridiculous. I'll let you appreciate what I mean. The indices are the benchmarks followed by the corresponding Vanguard funds (which are active). Tracking error is relative to the VIPSX data series (it's roughly an RMSE formula). Index series do not include any ER adjustment. Blue lines express my puzzlement.
Siamond,

I don't see what puzzles you. Is it the difference between annual returns and 10-year rates, or the difference between annual returns and annual inflation?


While waiting for your answer, here are some comments about intermediate bond funds.

Personally, I don't find the ITT annual returns puzzling, when compared to 10-year rates. They are pretty much explained by our Bond fund simulator's Bond Fund (10-4). Here are its returns:

Code: Select all

1997	11.48%
1998	9.57%
1999	-4.41%
2000	15.14%
2001	7.48%
2002	11.91%
2003	4.32%
2004	2.64%
2005	2.06%
2006	2.47%
2007	13.08%
2008	11.57%
2009	-2.48%
2010	7.05%
2011	11.43%
2012	2.89%
2013	-2.30%
2014	6.69%
2015	1.22%
There are small differences in annual returns between Bond Fund (10-4) and the ITT index, but this could easily be explained by the fact that our simulator uses "January average" yields to calculate its returns, and by the fact that it only uses annual yields, approximates missing yields at various maturities, and it is not market weighted. A growth chart shows how close the returns are:

Image

In other words, the annual returns are mostly explained by FRED's yield curve across time. A curious person can simply download the bond fund simulation spreadsheet and look at the detailed breakdown of each annual return. It is a very good exercise that helps understanding bond fund annual returns.

Bonds are not cash. They have a market value which varies across time. The nearest a bond is to maturity, the less its market value can differ from its par value. The farther a bond is to maturity, the more its market value can differ from its par value. Luckily, regardless of its value, the bond delivers its fixed coupons every year.

The long-term CAGR returns of a bond fund which keeps its bonds until maturity (or near to maturity) are mostly determined by the yields on bonds when they are acquired by the fund. If the fund buys all its bonds at par, long-term returns are mostly determined by coupons.

The same applies to TIPS. Many people naively assume that TIPS fund nominal returns should closely follow inflation. It's not so, because TIPS are bonds. This would be like expecting a nominal bond fund to have a fixed returns over time (to closely follow 0 inflation). For TIPS returns to closely follow inflation, the yield curve would have to be static (fixed, unchanging) across time, but this is practically impossible, because TIPS are traded on the market. Supply and demand set the price of traded TIPS. As real yields are a mirror of TIPS prices, it follows that supply and demand set real yields.

Someone seeking cash behavior (e.g. no variation in market value) from the "fixed income" allocation in his portfolio must invest into cash instruments. For stable nominal growth, CDs will do the job. For stable real growth, I-Bonds will do the job.

One should not expect an intermediate bond fund to behave like cash (or like inflation-indexed cash); this would be an unrealistic expectation. This applies as much to nominal intermediate Treasury funds as to intermediate TIPS funds.
Last edited by longinvest on Thu Apr 12, 2018 6:58 am, edited 3 times in total.
Variable Percentage Withdrawal (bogleheads.org/wiki/VPW) | One-Fund Portfolio (bogleheads.org/forum/viewtopic.php?t=287967)
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

longinvest wrote:I don't see what puzzles you. Is it the difference between annual returns and 10-year rates, or the difference between annual returns and annual inflation?
Sorry, should have been more explicit, and maybe I confused the point while adding too much data. (stlutz, I added STTs after seeing your post, by the way).

Compare TIPS and ITTs, either the corresponding real-life funds or the indices. I guess this truly makes your point that real yields have a life of their own compared to nominal yields. But the extent at which returns differs in many years (check the blue lines), in a time period of relatively mild inflation, is astounding to me. Yes, ok, average maturity is not perfectly aligned either, but still, this made my eyes pop. If you can find a reasonable explanation, I am all ears... :wink:
longinvest
Posts: 5672
Joined: Sat Aug 11, 2012 8:44 am

Re: How to model TIPS rates (and returns) pre-1997

Post by longinvest »

siamond wrote:
longinvest wrote:I don't see what puzzles you. Is it the difference between annual returns and 10-year rates, or the difference between annual returns and annual inflation?
Sorry, should have been more explicit, and maybe I confused the point while adding too much data. (stlutz, I added STTs after seeing your post, by the way).

Compare TIPS and ITTs, either the corresponding real-life funds or the indices. I guess this truly makes your point that real yields have a life of their own compared to nominal yields. But the extent at which this differs in many years (check the blue lines), in a time period of relatively mild inflation, is astounding to me. Yes, ok, average maturity is not perfectly aligned either, but still, this made my eyes pop. If you can find a reasonable explanation, I am all ears... :wink:
Oh! I fully agree. The significant differences between TIPS and ITT returns are just a mirror image of the apparent independence of nominal yields relative to real yields (which I called "inflation expectations") in the chart I posted earlier:

viewtopic.php?p=3159772#p3139955
longinvest wrote: Inflation Expectations (zero nominal coupon vs zero real coupon)
Image
From year to year, the differences are all over the map. Look at the difference between 2008 (very light blue, near the top) and 2009 (dark green, at the bottom). Even the shape of the expectation curve is significantly different.
Last edited by longinvest on Thu Apr 12, 2018 6:58 am, edited 1 time in total.
Variable Percentage Withdrawal (bogleheads.org/wiki/VPW) | One-Fund Portfolio (bogleheads.org/forum/viewtopic.php?t=287967)
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

dcabler wrote:Without arguing for the validity of having TIPs in either an accumulation or decumulation portfolio, my vote would be to keep a series in the spreadsheet along with the grain/block of salt disclaimer about the accuracy in the pre-1997 years.

siamond - I know at one time you were trying to get data for tips-equivalents from other countries in order to get more insight into how they have historically operated elsewhere. Any success there?
I did find the UK annual numbers for both regular bonds and TIPS, in the Barclays Equity Gilt study, a document updated every year. Well, the UK terminology is "gilts" and "inflation-linked gilts". Here is the link for the 2016 edition. You can easily copy and paste the numbers in a spreadsheet, and scrutinize to your heart's content. It would be great to have a good discussion about the UK track record at some point.

Then Larry Swedroe posted the following in a 2012 thread. I recently bumped the thread, asking him for data sources and pointers, but didn't elicit an answer. I didn't try to research the issue any further, but it would certainly be interesting to do so. Personally, I find the US track record quite aberrant (cf. my post last night), and would be glad to understand if this is an endemic problem or not.
larryswedroe wrote:While the Treasury issued its first inflation-protected security in 1997, it was not the first time they had been sold in the U.S. “The Commonwealth of Massachusetts, in 1780, issued the world’s first inflation-indexed bonds. The bonds were invented to deal with severe wartime inflation and with the angry discontent among soldiers in the U.S. Army with the decline in purchasing power of their pay. Although the bonds were successful, the concept of indexed bonds was abandoned after the immediate extreme inflationary environment passed, and largely forgotten until the twentieth century.”
Inflation-indexed bonds were reintroduced to the market, by Finland in 1945. Israel and Iceland were next in 1955. They were followed in the 1960s by Brazil, Chile, Columbia, and then in the early seventies by Argentina. Shortly thereafter the U.K. became the first developed market to introduce them. By the end of the century there were over twenty countries issuing inflation-linked bonds. Thus while the market for Treasury Inflation-Protected Securities is relatively new, we do have data on the performance (and risks) of these type bonds for longer periods.
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

Kevin M wrote:However, it would be nice to archive efforts like this in a way that future Bogleheads who might take up the mantle of maintaining and improving the backtest spreadsheet could learn from it and avoid going down blind alleys.
What I try to do is to document pointers to investigation/feedback threads like this one, plus short summaries about conclusions reached, in the main Simba thread. It proved very helpful for me to go through the entire Simba thread, and better understand the history, who did what and why. So I'm trying to extend this practice and provide proper traceability. In addition, I recently developed a private spreadsheet with MANY numbers, pointers, notes, comparison stats, etc, with one tab per Simba primary data series. This one can't be publicly shared, but I will definitely pass it on to the next volunteer.

PS. oh, and thank you for the nice words. Longinvest and Tyler definitely deserve special accolades.
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

stlutz wrote:My opinion remains to use the *short-term* Treasury series prior to 1997. Prior to then, if you wanted to guard your bond portfolio against inflation, this is what you used--that was the only actual option available at the time. With less duration risk, you were less exposed to increases in inflation.
Tyler9000 wrote:As for just using another bond type prior to 1997, I'm working on some tools to directly compare two series for how well they track one another over time. It would take too long to explain in a post like this, but the short story is that I agree that intermediate treasuries are a poor fit as a direct stand-in. Short term treasuries are really no better. I'll keep experimenting, but I'm not confident that will bear fruit with a reasonable amount of error to represent as the same thing.
dcabler wrote:Without arguing for the validity of having TIPs in either an accumulation or decumulation portfolio, my vote would be to keep a series in the spreadsheet along with the grain/block of salt disclaimer about the accuracy in the pre-1997 years.
Like dcabler, it really bugs me to have *nothing* to offer as a rough guess-timate for 1985-1996. Stlutz, maybe you're onto something (I like the pragmatic thinking here), but I agree with Tyler, STTs are a really poor fit for the 1997+ trajectory, which is troubling. And longinvest did convince me that the Kothari data (or something derived from Groen or UMich) is dubious at best.

Any other idea? I'd like to reach a conclusion before the end of the year, and the next Simba update.
longinvest
Posts: 5672
Joined: Sat Aug 11, 2012 8:44 am

Re: How to model TIPS rates (and returns) pre-1997

Post by longinvest »

siamond wrote:Any other idea? I'd like to reach a conclusion before the end of the year, and the next Simba update.
How about letting the spreadsheet user set a replacement security for missing years of an asset? This way, it becomes transparent and under the total responsibility of the user.

It could be a setting for each asset that doesn't go as far back as 1871. It would even work for things like TBM, where the official TBM numbers wouldn't include synthetic returns (it would include index returns, but no fund simulator returns). For some assets, the missing-years replacement setting could be pre-filled, such as our simulator's Bond Fund (10 to 2-year) for TBM, for example.

If done well, it could even work in chains. Let's say you have three assets: A (2000-2015), B (1970-2015), and C (1871-2015). You set "C" as B's missing-years replacement. You also set ''B" as A's missing-years replacement. As a result, in backtests, A extends back to 1871. It is composed of C's returns from 1871 to 1969, B's returns from 1970 to 1999, and its own returns from 2000 to 2015.
Variable Percentage Withdrawal (bogleheads.org/wiki/VPW) | One-Fund Portfolio (bogleheads.org/forum/viewtopic.php?t=287967)
dcabler
Posts: 4482
Joined: Wed Feb 19, 2014 10:30 am

Re: How to model TIPS rates (and returns) pre-1997

Post by dcabler »

longinvest wrote:
siamond wrote:Any other idea? I'd like to reach a conclusion before the end of the year, and the next Simba update.
How about letting the spreadsheet user set a replacement security for missing years of an asset? This way, it becomes transparent and under the total responsibility of the user.

It could be a setting for each asset that doesn't go as far back as 1871. It would even work for things like TBM, where the official TBM numbers wouldn't include synthetic returns (it would include index returns, but no fund simulator returns). For some assets, the missing-years replacement setting could be pre-filled, such as our simulator's Bond Fund (10 to 2-year) for TBM, for example.

If done well, it could even work in chains. Let's say you have three assets: A (2000-2015), B (1970-2015), and C (1871-2015). You set "C" as B's missing-years replacement. You also set ''B" as A's missing-years replacement. As a result, in backtests, A extends back to 1871. It is composed of C's returns from 1871 to 1969, B's returns from 1970 to 1999, and its own returns from 2000 to 2015.
I like this idea! Though, siamond, didn't you already add a feature to let one add any asset one wishes, along with the returns series? With that, one could cut-and-paste bits and pieces from assets already in Simba or from anywhere else for that matter. One thing I like to play with in my own spreadsheets is to create series using whatever index a particular mutual fund uses today, as if it had been using that index all along. Let's me see, for example, what things might have looked like if Vanguard had been using CRSP for its SCV fund since the beginning.
dcabler
Posts: 4482
Joined: Wed Feb 19, 2014 10:30 am

Re: How to model TIPS rates (and returns) pre-1997

Post by dcabler »

longinvest wrote:
siamond wrote:
longinvest wrote:I don't see what puzzles you. Is it the difference between annual returns and 10-year rates, or the difference between annual returns and annual inflation?
Sorry, should have been more explicit, and maybe I confused the point while adding too much data. (stlutz, I added STTs after seeing your post, by the way).

Compare TIPS and ITTs, either the corresponding real-life funds or the indices. I guess this truly makes your point that real yields have a life of their own compared to nominal yields. But the extent at which this differs in many years (check the blue lines), in a time period of relatively mild inflation, is astounding to me. Yes, ok, average maturity is not perfectly aligned either, but still, this made my eyes pop. If you can find a reasonable explanation, I am all ears... :wink:
Oh! I fully agree. The significant differences between TIPS and ITT returns are just a mirror image of the apparent independence of nominal yields relative to real yields (which I called "inflation expectations") in the chart I posted earlier:

viewtopic.php?p=3159772#p3139955
longinvest wrote: Inflation Expectations (zero nominal coupon vs zero real coupon)
Image
From year to year, the differences are all over the map. Look at the difference between 2008 (very light blue, near the top) and 2009 (dark green, at the bottom). Even the shape of the expectation curve is significantly different.
I ran across this paper earlier today which might help explain at least some of what you're seeing, though I'm not sure of the predictability of all of the factors they show (starting on page 6)..

http://pages.stern.nyu.edu/~dbackus/GE_ ... g%2006.pdf
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

longinvest wrote:How about letting the spreadsheet user set a replacement security for missing years of an asset? This way, it becomes transparent and under the total responsibility of the user.
dcabler wrote:Though, siamond, didn't you already add a feature to let one add any asset one wishes, along with the returns series?
Not really a feature, but I added instructions explaining how to do so, and I tweaked various formulas here and there to make it a reasonably easy process when you know what you're doing.

This kind of stuff is for advanced users though. I still would like to have a set of default values to plug in... We have a few coarse approximations here and there to plug holes, and I have a note to myself to mark them more clearly, this would be one such case.

PS. Without stealing his thunder, Tyler is working on something closer to what longinvest suggested, including clever metrics to assess if one choice or another gets you in muddy waters with your guesswork.
dcabler
Posts: 4482
Joined: Wed Feb 19, 2014 10:30 am

Re: How to model TIPS rates (and returns) pre-1997

Post by dcabler »

siamond wrote:
longinvest wrote:How about letting the spreadsheet user set a replacement security for missing years of an asset? This way, it becomes transparent and under the total responsibility of the user.
dcabler wrote:Though, siamond, didn't you already add a feature to let one add any asset one wishes, along with the returns series?
Not really a feature, but I added instructions explaining how to do so, and I tweaked various formulas here and there to make it a reasonably easy process when you know what you're doing.

This kind of stuff is for advanced users though. I still would like to have a set of default values to plug in... We have a few coarse approximations here and there to plug holes, and I have a note to myself to mark them more clearly, this would be one such case.

PS. Without stealing his thunder, Tyler is working on something closer to what longinvest suggested, including clever metrics to assess if one choice or another gets you in muddy waters with your guesswork.
Ah, OK. Would still be cool to have adding ones own series in. And best to do it in such a way that it doesn't need to be completely recreated each time Simba is updated (maybe import from a text file). Perhaps Simba 2018? :D

I'd still like to vote to keep the full TIPs pseudo-sequence in all the way back to 70 or 72 instead of pushing it to 85 onwards. Would like to hold out some hope for an eventual good-enough sequence some day. :shock:
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

dcabler wrote:Ah, OK. Would still be cool to have adding ones own series in. And best to do it in such a way that it doesn't need to be completely recreated each time Simba is updated (maybe import from a text file). Perhaps Simba 2018?
I had a vague note in this respect, let me better capture such thoughts, so that I mull it over at some point. I do agree on the spirit of your suggestion.
dcabler wrote:I'd still like to vote to keep the full TIPs pseudo-sequence in all the way back to 70 or 72 instead of pushing it to 85 onwards. Would like to hold out some hope for an eventual good-enough sequence some day. :shock:
What I am going to do in any case is to provide the full Kothari series in the Data_Misc tab. But I do agree with longinvest that, at this stage, this is just too misleading to provide as 1970+ default in the primary series. Plus we really need to make 'naive' Simba users realize that TIPS are fairly new in the US, with a limited track record. In parallel, if somehow we could find a way to extend the Kothari series (or others) towards recent years and check it against actuals, this could create a better sense of comfort (or discomfort) to possibly revisit this cautious position.
=> Would that be a decent compromise?
dcabler
Posts: 4482
Joined: Wed Feb 19, 2014 10:30 am

Re: How to model TIPS rates (and returns) pre-1997

Post by dcabler »

siamond wrote:
dcabler wrote:Ah, OK. Would still be cool to have adding ones own series in. And best to do it in such a way that it doesn't need to be completely recreated each time Simba is updated (maybe import from a text file). Perhaps Simba 2018?
I had a vague note in this respect, let me better capture such thoughts, so that I mull it over at some point. I do agree on the spirit of your suggestion.
dcabler wrote:I'd still like to vote to keep the full TIPs pseudo-sequence in all the way back to 70 or 72 instead of pushing it to 85 onwards. Would like to hold out some hope for an eventual good-enough sequence some day. :shock:
What I am going to do in any case is to provide the full Kothari series in the Data_Misc tab. But I do agree with longinvest that, at this stage, this is just too misleading to provide as 1970+ default in the primary series. Plus we really need to make 'naive' Simba users realize that TIPS are fairly new in the US, with a limited track record. In parallel, if somehow we could find a way to extend the Kothari series (or others) towards recent years and check it against actuals, this could create a better sense of comfort (or discomfort) to possibly revisit this cautious position.
=> Would that be a decent compromise?
No problem at all. Plus as noted before, since you have instructions added on how to put in ones own series, that option is always there to play with.
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

dcabler alerted me to a useful pointer (thank you!). Barclays provides daily time series of TIPS indices, available for download. For example:
US Govt Inflation-Linked (BCIT1T)
UK Govt Inflation-Linked (BCIU1T)
Barclays/ABSA South Africa Government Inflation-Linked (BEMZ0Z)

Unfortunately, this is mostly it (besides some aggregate indices), I didn't find older data series like Finland, Israel, Iceland, etc.

As a side note, the truly passionate may want to take a look at this book: Inflation-indexed Securities: Bonds, Swaps and Other Derivatives by Mark Deacon and al. It is pretty expensive, but generous excerpts about International TIPS history (covering many countries) can be found on Google Books (click here, then search for other indexed bond markets).
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

siamond wrote:What I am going to do in any case is to provide the full Kothari series in the Data_Misc tab.
Ok, done. The last update of the Simba spreadsheet (working version, temporary file) includes the 1954-1998 data series derived from the Kothari/Shanken research in the Data_Misc tab, plus corresponding references (and a pointer to this thread). The primary series (former S-TIPS) is now named 'TIPS', and restricted to 1985+ returns, in the Data_TR_USD. The 1985-1996 numbers still come from the Kothari/Shanken series while we finish mulling over what to do for this time period. In the process, I also cleaned up the separate spreadsheet where I did the annual math, derived from the original (monthly, TTM) series from the professors.
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

I had some difficulties believing that IT Treasuries would behave so differently from TIPS in the 1997+ time period, as this table of annual returns seemed to indicate. I started to wonder if this was more an issue with year boundaries than a really dramatic divergence. I ended up downloading monthly returns of the corresponding indices (also including ST Treasuries while I was at it), and created a growth chart and a telltale chart (relative to the TIPS series), hence cumulative quantities less sensitive to date boundaries.

Here are the indices I used:
- Bloomberg Barclays US Treasury US TIPS TR USD
- Bloomberg Barclays 1-5 Yr Treasury TR USD
- Bloomberg Barclays 5-10 Yr Treasury TR USD

Image

Seems to me that, after all, IT Treasuries might not be such a bad (crude) proxy for TIPS for the 1985-1996 period (while inflation was reasonably tame, and no big crisis occurred). Checking some numbers on the monthly series, CAGR and tracking error do go back to very reasonable values (as should be obvious from looking at the charts). Thoughts?
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

I also wanted to do a test with a better match about average maturity. Unfortunately, I can't find a Barclays 5-10Y TIPS series. A 1-5Y TIPS series is available since Dec-04 though. So I compared the following:
- Bloomberg Barclays U.S. Treasury TIPS 1-5Y TR USD (used as reference for the telltale chart)
- Bloomberg Barclays 1-5 Yr Treasury TR USD
- Bloomberg Barclays US Treasury US TIPS TR USD

Ok, so aside from the liquidity hiccup of Q4-08, this all lines up pretty well. Which reinforces me in my previous proposal. During a time of fairly mild (hence predictable) inflation, using IT Treasuries appears to be a decent proxy for TIPS historical returns.

Image
User avatar
Tyler9000
Posts: 739
Joined: Fri Aug 21, 2015 11:57 am

Re: How to model TIPS rates (and returns) pre-1997

Post by Tyler9000 »

siamond wrote: During a time of fairly mild (hence predictable) inflation, using IT Treasuries appears to be a decent proxy for TIPS historical returns.
That makes logical sense, so it's good to see the data match that assumption. I also agree with your decision to NOT extend it back pre-1985. With high inflation, I would expect them to behave very differently (as designed).
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

Here is the final decision about the new TIPS series in the Simba update to come early Jan-17:

- Longinvest's bonds fund spreadsheet (10-4 model) 1985-1991
- Bloomberg Barclays 5-10 Yr Treasury TR USD 1992-1996
- Bloomberg Barclays US TIPS index (Series L) TR USD 1997-2000
- Vanguard Inflation-Protected Security Fund (VIPSX) 2001+

Many thanks for all the great discussion and modeling attempts. We reached a temporary conclusion for now, but clearly, more research would definitely be welcome on this topic, and maybe we'll find a better way later on.
MachineGhost
Posts: 108
Joined: Sat Feb 14, 2009 1:46 am

Re: How to model TIPS rates (and returns) pre-1997

Post by MachineGhost »

siamond wrote:Many thanks for all the great discussion and modeling attempts. We reached a temporary conclusion for now, but clearly, more research would definitely be welcome on this topic, and maybe we'll find a better way later on.
Here's a thought. Why not model EE-Bonds or I-Bonds? They are relatively simple to do so as they don't rely on inflation expectations, just CPI increase + a fixed rate. There may be a way to determine a dynamic fixed rate easier than synthesizing a yield curve TIPS.

Inflation expectations is definitely a strange beast to try and lasso. Another thought: It may be possible to extract it from the behavior of commodities or other real assets.

I don't know if its well known around here, but inflation expectations during the late 60's and early 70's were consistently understating actual inflation even while inflation was consistently increasing (talk about a long-term anchoring bias!). Doom porn tent revivalists were out shouting as early as 1965 after silver coin convertibility was closed so its not like the later "inflation shock" was a true Black Swan. The smart money was prepared long before Nixon closed the gold window. The inflation expectations sentiment only shifted to overstating during the late 70's to early 80's and persists to date. Said another way, I vote for the UMich in having possible merit since it should reflect the dumb money and they ultimately drive inflation expectations en masse.
AlohaJoe
Posts: 6609
Joined: Mon Nov 26, 2007 1:00 pm
Location: Saigon, Vietnam

Re: How to model TIPS rates (and returns) pre-1997

Post by AlohaJoe »

I don't actually remember what the blocker for this was. But I was reading something else recently that mentioned a 2008 paper that is maybe related.

In 2008 Ang, Bekaert, and Wei published "The Term Structure of Real Rates and Expected Inflation" where they create a model for real rates (both short & long) as well as expected inflation going back to 1950.

Dunno if it is useful or not.

Since Ang is relatively well known (he's now the Head of Factor blah blah something at Black Rock; they poached him academic in 2015), there's also a bunch of follow up research along the same lines.
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

MachineGhost wrote:Here's a thought. Why not model EE-Bonds or I-Bonds? They are relatively simple to do so as they don't rely on inflation expectations, just CPI increase + a fixed rate. There may be a way to determine a dynamic fixed rate easier than synthesizing a yield curve TIPS.
AlohaJoe wrote:I don't actually remember what the blocker for this was.
Well, this would be another project. The point of this thread was to try to identify a decent way to model TIPS to extend the corresponding series in Simba back in time. As this thread developed, I got the clear impression that this really can't be done with any chance of reasonable accuracy, as TIPS pricing is a direct function of 'sentiment' about expected inflation (and other factors like liquidity), and this is the kind of stuff we really can't reconstruct back in time. Inflation proved highly unpredictable, and the same undoubtedly applies to the 'sentiment' about expected inflation...

As to EE-Bonds or I-Bonds, this would be another project. EE-Bonds are very deterministic unless one sells prematurely, so I'm not sure modeling those would be terribly valuable. I could see value in modeling I-Bonds if there is interest, and yes, this sounds much simpler to do. They were created in 1998, I believe (see here), so this should give enough years to validate a model, and then backtrack. Any taker?
User avatar
Tyler9000
Posts: 739
Joined: Fri Aug 21, 2015 11:57 am

Re: How to model TIPS rates (and returns) pre-1997

Post by Tyler9000 »

AlohaJoe wrote: In 2008 Ang, Bekaert, and Wei published "The Term Structure of Real Rates and Expected Inflation" where they create a model for real rates (both short & long) as well as expected inflation going back to 1950.

Dunno if it is useful or not.
Nice find! The paper itself seems short on verifiable data, but perhaps there's something useful in the references. I'm skeptical, but I'll dig a bit and see what I can find.
AlohaJoe
Posts: 6609
Joined: Mon Nov 26, 2007 1:00 pm
Location: Saigon, Vietnam

Re: How to model TIPS rates (and returns) pre-1997

Post by AlohaJoe »

Tyler9000 wrote:
AlohaJoe wrote: In 2008 Ang, Bekaert, and Wei published "The Term Structure of Real Rates and Expected Inflation" where they create a model for real rates (both short & long) as well as expected inflation going back to 1950.

Dunno if it is useful or not.
Nice find! The paper itself seems short on verifiable data, but perhaps there's something useful in the references. I'm skeptical, but I'll dig a bit and see what I can find.
It is possible Ang has more detailed data available. Perhaps a spreadsheet from which the charts were derived. I emailed him once in the past and he replied quickly, so it is possible he might be able to help.
rgordon434
Posts: 1
Joined: Mon Jun 26, 2017 8:56 am

Re: How to model TIPS rates (and returns) pre-1997

Post by rgordon434 »

grok87 wrote:
longinvest wrote:
grok87 wrote: I haven't read through the entire "Historical Bond Returns - Shiller: From Rates to Returns" thread yet but i will. I think the key assumption is to fill in the 19 year TIPS real yields and the 9 year TIPs real yields for each year. The key assumption is the local slope of the yield curve around the 20 year and 10 year points. It seems like it should be possible to build on the nominal yield curve slopes that were used in the previous analysis- ie the nominal and real yield curve slopes are probably not TOO different...

I'll copy in the folks on the other thread...
Grok,

Please take the time to entirely read the first post of the "Historival Bond Returns..." thread. It explains important concepts behind its model.

longinvest
THanks longinvest,
I just read the first post and i opened up your spreadsheet. Looks like a lot of good work has been done already and we don't want to reinvent the wheel. Can you just plug in the 20 year real rates and the 10 year real rates from the New York Fed (Groen and Middledorp) paper
http://libertystreeteconomics.newyorkfe ... tions.html
(spreadsheet link a the bottom of the paper)
and come up with the annual real returns for a long term tips fund: 10-20 year maturity I guess?
cheers,
Hey Grok,

Did we ever reach a conclusion on loading the bond fund simulator up with the 10-20 year yields derived from the regression model developed by
New York Fed / Groen & Middeldorp? Let me know what you think. It seems like that would be a meaningful exercise.

I am in the process of building an asset allocation and would love to be able to backtest to the 1970s. I am new on Bogelhead. You all are doing a great service to the investment community.

Best,

Rob
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

rgordon434 wrote:Hey Grok,

Did we ever reach a conclusion on loading the bond fund simulator up with the 10-20 year yields derived from the regression model developed by
New York Fed / Groen & Middeldorp? Let me know what you think. It seems like that would be a meaningful exercise.
Hi Rob. I am not Grok87, but I can tell you that several of us tried to look to this model from Groen & Middeldorp, and couldn't convince ourselves that it was solid enough for our needs (i.e. to feed the simulator). I actually tried to contact both authors to get more details, so that we can try to better validate it against known numbers, but alas, didn't get any response.

Overall, I know this is not satisfying at all, but we had to acknowledge defeat (at least for now), and in the Simba backtesting spreadsheet, we restricted historical TIPS returns to 1985+, because we just couldn't find anything that seems trustworthy enough for previous years.
AlohaJoe
Posts: 6609
Joined: Mon Nov 26, 2007 1:00 pm
Location: Saigon, Vietnam

Re: How to model TIPS rates (and returns) pre-1997

Post by AlohaJoe »

I stumbled across a September 2018 paper from the Journal of Empirical Finance that also tries to simulate TIPS returns. "Simulating historical inflation-linked bond returns" by Laurens Swinkels. He claims it is "more accurate and less data intensive" than Kothari's approach. He uses his method to simulate TIPS for 41 countries from 1987-2017.

The paper claims that the full data series are available online but I couldn't find them.

https://www.sciencedirect.com/science/a ... via%3Dihub

Related: the same author has (simulated) monthly bond returns for US (1947-), Germany (1972-), Japan (1974-), Australia (1969-), Norway (1921-), and Sweden (1920-).

https://eur.figshare.com/articles/Data_ ... 62/8152748
dcabler
Posts: 4482
Joined: Wed Feb 19, 2014 10:30 am

Re: How to model TIPS rates (and returns) pre-1997

Post by dcabler »

AlohaJoe wrote: Sun Dec 15, 2019 2:21 am I stumbled across a September 2018 paper from the Journal of Empirical Finance that also tries to simulate TIPS returns. "Simulating historical inflation-linked bond returns" by Laurens Swinkels. He claims it is "more accurate and less data intensive" than Kothari's approach. He uses his method to simulate TIPS for 41 countries from 1987-2017.

The paper claims that the full data series are available online but I couldn't find them.

https://www.sciencedirect.com/science/a ... via%3Dihub

Related: the same author has (simulated) monthly bond returns for US (1947-), Germany (1972-), Japan (1974-), Australia (1969-), Norway (1921-), and Sweden (1920-).

https://eur.figshare.com/articles/Data_ ... 62/8152748
Here's another link to download the paper that doesn't require establishing any kind of account ( at least for your first shot at it) :
https://papers.ssrn.com/sol3/papers.cfm ... id=2894869

Couldn't find the data either, might need to ask.

Embedded in the paper is a link to another study, with excel data, I hadn't seen before from Chernov & Mueller

Finally, I wonder why the author didn't compare to the Liberty Street Paper and data - it's even more data intensive than Kothari's work.
https://libertystreeteconomics.newyorkf ... tions.html

Cheers
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

AlohaJoe wrote: Sun Dec 15, 2019 2:21 am I stumbled across a September 2018 paper from the Journal of Empirical Finance that also tries to simulate TIPS returns. "Simulating historical inflation-linked bond returns" by Laurens Swinkels. He claims it is "more accurate and less data intensive" than Kothari's approach. He uses his method to simulate TIPS for 41 countries from 1987-2017.
This seems like a well researched and remarkably honest paper, thank you for sharing. It is important to appreciate how approximate such a TIPS model is, even when refined compared to previous work (a correlation of 0.7 or 0.8 -not R2, just straight correlation- with actuals isn't that great, as acknowledged by the authors). As we tentatively concluded a while ago on this forum, the problem of simulating historical TIPS with solid accuracy seems pretty much intractable...

As a side note, this other article from the same author looks quite interesting too (Historical Data: International monthly government bond returns). And this one does come with an easy to download data set.
AlohaJoe
Posts: 6609
Joined: Mon Nov 26, 2007 1:00 pm
Location: Saigon, Vietnam

Re: How to model TIPS rates (and returns) pre-1997

Post by AlohaJoe »

siamond wrote: Sun Dec 15, 2019 11:55 am As we tentatively concluded a while ago on this forum, the problem of simulating historical TIPS with solid accuracy seems pretty much intractable...
Yep, I probably should have mentioned that in my post. He's pretty clear about the limited usefulness of the simulations. Too bad he didn't cite Bogleheads :D
User avatar
Topic Author
siamond
Posts: 6003
Joined: Mon May 28, 2012 5:50 am

Re: How to model TIPS rates (and returns) pre-1997

Post by siamond »

AlohaJoe wrote: Sun Dec 15, 2019 6:55 pm
siamond wrote: Sun Dec 15, 2019 11:55 am As we tentatively concluded a while ago on this forum, the problem of simulating historical TIPS with solid accuracy seems pretty much intractable...
Yep, I probably should have mentioned that in my post. He's pretty clear about the limited usefulness of the simulations. Too bad he didn't cite Bogleheads :D
Thinking more about it, I would love to have access to the corresponding data (for the US) and give it a good comparison with the Barclays TIPS index while adding to the chart the Barclays 5-10 Yr Treasuries index. Even if correlation numbers aren't overly impressive, this might still be a solid improvement over the atrociously crude mapping we currently have in Simba for the 1987-1996 decade (i.e. the 5-10 Yr Treasuries index). If the author is willing to share (assuming a proper citation, of course!), one possibility would be to provide this new model as an option in the new 'splicing' format of the Simba spreadsheet (to be officialized in Jan-2020).

As a side note, I find it really impressive that the author ran all those out-of-sample tests in numerous countries. Very solid research and a lot of hard work for sure.
dcabler
Posts: 4482
Joined: Wed Feb 19, 2014 10:30 am

Re: How to model TIPS rates (and returns) pre-1997

Post by dcabler »

siamond wrote: Tue Dec 17, 2019 9:07 am
AlohaJoe wrote: Sun Dec 15, 2019 6:55 pm
siamond wrote: Sun Dec 15, 2019 11:55 am As we tentatively concluded a while ago on this forum, the problem of simulating historical TIPS with solid accuracy seems pretty much intractable...
Yep, I probably should have mentioned that in my post. He's pretty clear about the limited usefulness of the simulations. Too bad he didn't cite Bogleheads :D
Thinking more about it, I would love to have access to the corresponding data (for the US) and give it a good comparison with the Barclays TIPS index while adding to the chart the Barclays 5-10 Yr Treasuries index. Even if correlation numbers aren't overly impressive, this might still be a solid improvement over the atrociously crude mapping we currently have in Simba for the 1987-1996 decade (i.e. the 5-10 Yr Treasuries index). If the author is willing to share (assuming a proper citation, of course!), one possibility would be to provide this new model as an option in the new 'splicing' format of the Simba spreadsheet (to be officialized in Jan-2020).

As a side note, I find it really impressive that the author ran all those out-of-sample tests in numerous countries. Very solid research and a lot of hard work for sure.
Indeed! FYI - I am currently in contact with the author and I expect that the data will be available within 1-2 days. Stay tuned! I'll post the info.
Cheers
Post Reply