[Thrift Savings Plan - Fund composition and comparisons]

Discuss all general (i.e. non-personal) investing questions and issues, investing news, and theory.
Post Reply
User avatar
Topic Author
vineviz
Posts: 14921
Joined: Tue May 15, 2018 1:55 pm
Location: Baltimore, MD

[Thrift Savings Plan - Fund composition and comparisons]

Post by vineviz »

[Split into a new thread from: Improving the TSP [for current participants] --admin LadyGeek]
tadamsmar wrote: Mon Oct 07, 2019 7:59 am And, unlike a bond fund, it never loses money in nominal dollars (it can lose money to inflation if the inflation rate is higher than the interest rate).
This is a virtue for portfolios which demand almost no volatility. For more typical portfolios, the "never loses money in nominal dollars" is a bug not a feature because it makes the G Fund completely useless at providing diversification.
tadamsmar wrote: Mon Oct 07, 2019 7:59 am The efficient frontier calculation is not clearly faulty. It just uses different assumptions than you did.
The only way to produce the allocations in the L Fund portfolios is to assume that the expected returns of the G Fund and F Fund are precisely equal, which is a fairly aggressive assumption.
"Far more money has been lost by investors preparing for corrections than has been lost in corrections themselves." ~~ Peter Lynch
rkhusky
Posts: 17659
Joined: Thu Aug 18, 2011 8:09 pm

Re: Improving the TSP [for current participants]

Post by rkhusky »

vineviz wrote: Mon Oct 07, 2019 9:57 am The only way to produce the allocations in the L Fund portfolios is to assume that the expected returns of the G Fund and F Fund are precisely equal, which is a fairly aggressive assumption.
The Treasuries in the F Fund should have approximately the same yield as the G Fund. The corporate bonds in the F Fund should have higher yield due to higher risk. The returns of the F Fund will also differ from the G Fund due to interest rate changes, but those changes should approximately cancel out over time. Any differences due to interest changes will only be 2nd or 3rd order effects and will be necessarily small.

The risk of the corporate bonds will just move the F Fund and G Fund to different spots on the efficient frontier. Any effects of interest rate changes will cause only small differences, if any, in the efficiencies of the F Fund and G Fund. Most people don't really care about small fractions of a percent in efficiencies and will instead choose F based on the expected out-performance over G, which is primarily due to the corporate bonds, while others will choose G for its inherent safety.
User avatar
tadamsmar
Posts: 9972
Joined: Mon May 07, 2007 12:33 pm

Re: Improving the TSP [for current participants]

Post by tadamsmar »

vineviz wrote: Mon Oct 07, 2019 9:57 am
tadamsmar wrote: Mon Oct 07, 2019 7:59 am And, unlike a bond fund, it never loses money in nominal dollars (it can lose money to inflation if the inflation rate is higher than the interest rate).
This is a virtue for portfolios which demand almost no volatility. For more typical portfolios, the "never loses money in nominal dollars" is a bug not a feature because it makes the G Fund completely useless at providing diversification.
I don't think you are properly using the term "diversification". I think you mean you you get less negative correlation with G vs F, that's true. But the whole point in the efficient frontier calculation is to take all this into account and provide optimal risk-adjusted return at each level of risk. Do you understand that bad-mouthing the specific characteristics of the assets involved in this calculation makes no difference because the calculation itself accounts for all that?

Have you noticed that on the low end of the risk domain that the efficient frontier calculation allocates some money to the asset that represents the risk-free rate? Yet, that asset provides no diversification (according to your definition of "diversification"). If you think that is a bad thing, then logically you should reject the whole concept of using the efficient frontier since this is an inherent characteristic of the calculation.
tadamsmar wrote: Mon Oct 07, 2019 7:59 am The efficient frontier calculation is not clearly faulty. It just uses different assumptions than you did.
The only way to produce the allocations in the L Fund portfolios is to assume that the expected returns of the G Fund and F Fund are precisely equal, which is a fairly aggressive assumption.
Perhaps I don't understand what you are saying, it seems to be obviously untrue since the L Fund estimation already assumes that G and F are not equal.
User avatar
tadamsmar
Posts: 9972
Joined: Mon May 07, 2007 12:33 pm

Re: Improving the TSP [for current participants]

Post by tadamsmar »

vineviz wrote: Mon Oct 07, 2019 9:57 am
tadamsmar wrote: Mon Oct 07, 2019 7:59 am And, unlike a bond fund, it never loses money in nominal dollars (it can lose money to inflation if the inflation rate is higher than the interest rate).
This is a virtue for portfolios which demand almost no volatility. For more typical portfolios, the "never loses money in nominal dollars" is a bug not a feature because it makes the G Fund completely useless at providing diversification.
tadamsmar wrote: Mon Oct 07, 2019 7:59 am The efficient frontier calculation is not clearly faulty. It just uses different assumptions than you did.
The only way to produce the allocations in the L Fund portfolios is to assume that the expected returns of the G Fund and F Fund are precisely equal, which is a fairly aggressive assumption.
Maybe this will help you understand how the G Fund impacts the L Fund efficient frontier:

Q16: Why does the L Funds' efficient frontier line look so flat? A: The efficient frontier is usually described by a curved line. However, in the case of the L Funds, the curve of the line is affected by the unique risk and return characteristics of the G Fund. The G Fund is able to deliver rates of return like those on long-term Treasury securities with very little expected variance of returns (i.e., little risk). This means that expected returns on the left end of the line (where the G Fund is the most significant part of the L Fund portfolios) are higher for the amount of risk taken than would normally be the case. This results in a flatter line.
This is from a TSP FAQ that is no longer available, but the point made is true.

You can see this efficient frontier here:

https://www.tsp.gov/PDF/formspubs/FundsL.pdf

Also, that link makes it clear that the L Fund is based on quantitative analysis, not fund popularity:
The asset allocations are based on the investment consultant’s assumptions regarding future investment returns, inflation, economic growth, and interest rates.
The collection of optimal asset allocations make up the “Efficient Frontier,” which is shown by the curve. Asset allocations that are below the Efficient Frontier are less than optimal, because there is an asset allocation along the frontier that has a higher expected return for the same level of risk, or lower risk for the same expected return.
Another way to understand this is to do a thought experiment. Suppose you had access to a risk-free asset that had an estimated 20% return. If you calculated an efficient frontier with that asset along with TBM and TSM then you would have a 100% allocation to the risk-free asset at all risk levels. A high yielding risk-free asset can have a profound effect on the efficient frontier. The G Fund is not completely risk-free because of the possibility of unanticipated inflation that is not reflected in the bond rates, but it has a similar effect on the efficient frontier.

The G Fund inception was 1987 and bond rates, the 10-year T-bill for instance, have been trending down since that date and that makes the trailing F Fund look relatively good since the inception date. So if you just use the available trailing returns of G and F then you will get a high allocation to F. But I am sure that the TSP contractor does not just use this biased data. I am sure they do some kind of longer-term modeling.
User avatar
Topic Author
vineviz
Posts: 14921
Joined: Tue May 15, 2018 1:55 pm
Location: Baltimore, MD

Re: Improving the TSP [for current participants]

Post by vineviz »

tadamsmar wrote: Mon Oct 07, 2019 8:09 pm I don't think you are properly using the term "diversification". I think you mean you you get less negative correlation with G vs F, that's true. But the whole point in the efficient frontier calculation is to take all this into account and provide optimal risk-adjusted return at each level of risk.
It might seem like a semantic point, but the efficient frontier used to construct the L funds is making assumptions about expected return and expected risk. In other words, it's only "optimal" given that the assumptions hold true. And the only way the plan could make the allocations they've made is to assume that future returns of the G Fund and F Fund are identical.

If you think that assumption is reasonable, then you should be content that the L Funds are allocating between G and F appropriately.

I don't think that's a very reasonable assumption, given the composition of the two funds, but YMMV.
"Far more money has been lost by investors preparing for corrections than has been lost in corrections themselves." ~~ Peter Lynch
User avatar
tadamsmar
Posts: 9972
Joined: Mon May 07, 2007 12:33 pm

Re: Improving the TSP [for current participants]

Post by tadamsmar »

vineviz wrote: Tue Oct 08, 2019 9:31 am
tadamsmar wrote: Mon Oct 07, 2019 8:09 pm I don't think you are properly using the term "diversification". I think you mean you you get less negative correlation with G vs F, that's true. But the whole point in the efficient frontier calculation is to take all this into account and provide optimal risk-adjusted return at each level of risk.
It might seem like a semantic point, but the efficient frontier used to construct the L funds is making assumptions about expected return and expected risk. In other words, it's only "optimal" given that the assumptions hold true. And the only way the plan could make the allocations they've made is to assume that future returns of the G Fund and F Fund are identical.

If you think that assumption is reasonable, then you should be content that the L Funds are allocating between G and F appropriately.

I don't think that's a very reasonable assumption, given the composition of the two funds, but YMMV.
They are probably assuming F has a somewhat higher return than G. Seems to me that is a reasonable assumption and it looks to be consistent with the facts that we know. The allocations depend on the correlation matrices of the 5 TSP assets.

I think the assumptions are reasonable. As I said earlier, it's fuzzy math. The guy who invented the efficient frontier never used it for his personal nest egg, he just allocated 50-50 to stocks and bonds!
User avatar
Topic Author
vineviz
Posts: 14921
Joined: Tue May 15, 2018 1:55 pm
Location: Baltimore, MD

Re: Improving the TSP [for current participants]

Post by vineviz »

tadamsmar wrote: Tue Oct 08, 2019 11:06 am They are probably assuming F has a somewhat higher return than G. Seems to me that is a reasonable assumption and it looks to be consistent with the facts that we know. The allocations depend on the correlation matrices of the 5 TSP assets.
Sorry, but no. The only way to generate an efficient frontier which contains the L fund allocations is to assume that the two funds have the same expected return. If the consultants had assumed that "F has a somewhat higher return than G", the L funds would have much lower allocations to the G Fund than they actually do.
"Far more money has been lost by investors preparing for corrections than has been lost in corrections themselves." ~~ Peter Lynch
rkhusky
Posts: 17659
Joined: Thu Aug 18, 2011 8:09 pm

Re: Improving the TSP [for current participants]

Post by rkhusky »

The efficient frontier calculation also involves volatility.
trueblueky
Posts: 2365
Joined: Tue May 27, 2014 3:50 pm

Re: Improving the TSP [for current participants]

Post by trueblueky »

vineviz wrote: Tue Oct 08, 2019 12:30 pm
tadamsmar wrote: Tue Oct 08, 2019 11:06 am They are probably assuming F has a somewhat higher return than G. Seems to me that is a reasonable assumption and it looks to be consistent with the facts that we know. The allocations depend on the correlation matrices of the 5 TSP assets.
Sorry, but no. The only way to generate an efficient frontier which contains the L fund allocations is to assume that the two funds have the same expected return. If the consultants had assumed that "F has a somewhat higher return than G", the L funds would have much lower allocations to the G Fund than they actually do.
Is it the same risk-adjusted return?
User avatar
Topic Author
vineviz
Posts: 14921
Joined: Tue May 15, 2018 1:55 pm
Location: Baltimore, MD

Re: Improving the TSP [for current participants]

Post by vineviz »

trueblueky wrote: Tue Oct 08, 2019 2:28 pm
vineviz wrote: Tue Oct 08, 2019 12:30 pm
tadamsmar wrote: Tue Oct 08, 2019 11:06 am They are probably assuming F has a somewhat higher return than G. Seems to me that is a reasonable assumption and it looks to be consistent with the facts that we know. The allocations depend on the correlation matrices of the 5 TSP assets.
Sorry, but no. The only way to generate an efficient frontier which contains the L fund allocations is to assume that the two funds have the same expected return. If the consultants had assumed that "F has a somewhat higher return than G", the L funds would have much lower allocations to the G Fund than they actually do.
Is it the same risk-adjusted return?
The G Fund has no virtually no price volatility (it can't lose value, for instance) so it's arguably not very useful to think about the risk-adjusted return of that asset in isolation. But under the assumption that the F Fund and G Fund have the same expected returns, a 60/40 portfolio (just to take an example) would have roughly the same risk-adjusted return - Sharpe ratios identical to the second decimal - regardless of how the 40% was allocated between F and G.
"Far more money has been lost by investors preparing for corrections than has been lost in corrections themselves." ~~ Peter Lynch
User avatar
tadamsmar
Posts: 9972
Joined: Mon May 07, 2007 12:33 pm

Re: Improving the TSP [for current participants]

Post by tadamsmar »

vineviz wrote: Tue Oct 08, 2019 12:30 pm
tadamsmar wrote: Tue Oct 08, 2019 11:06 am They are probably assuming F has a somewhat higher return than G. Seems to me that is a reasonable assumption and it looks to be consistent with the facts that we know. The allocations depend on the correlation matrices of the 5 TSP assets.
Sorry, but no. The only way to generate an efficient frontier which contains the L fund allocations is to assume that the two funds have the same expected return. If the consultants had assumed that "F has a somewhat higher return than G", the L funds would have much lower allocations to the G Fund than they actually do.
Kindly provide a mathematical proof for that mathematical proposition.
User avatar
Topic Author
vineviz
Posts: 14921
Joined: Tue May 15, 2018 1:55 pm
Location: Baltimore, MD

Re: Improving the TSP [for current participants]

Post by vineviz »

tadamsmar wrote: Tue Oct 08, 2019 5:15 pm Kindly provide a mathematical proof for that mathematical proposition.
It's not a "mathematical proposition" so much as an empirical observation: the only solution to the Markowitz equations which generates an efficient frontier that contains the L fund allocations is one in which the expected return of the G Fund and F Funds are identical.

If you don't believe me, try it yourself and see what you get.
"Far more money has been lost by investors preparing for corrections than has been lost in corrections themselves." ~~ Peter Lynch
User avatar
tadamsmar
Posts: 9972
Joined: Mon May 07, 2007 12:33 pm

Re: Improving the TSP [for current participants]

Post by tadamsmar »

vineviz wrote: Tue Oct 08, 2019 5:42 pm
tadamsmar wrote: Tue Oct 08, 2019 5:15 pm Kindly provide a mathematical proof for that mathematical proposition.
It's not a "mathematical proposition" so much as an empirical observation: the only solution to the Markowitz equations which generates an efficient frontier that contains the L fund allocations is one in which the expected return of the G Fund and F Funds are identical.

If you don't believe me, try it yourself and see what you get.
But the return is not the only factor that determines the allocations. Therefore, your claim is false.
User avatar
Topic Author
vineviz
Posts: 14921
Joined: Tue May 15, 2018 1:55 pm
Location: Baltimore, MD

Re: Improving the TSP [for current participants]

Post by vineviz »

tadamsmar wrote: Tue Oct 08, 2019 8:31 pm
vineviz wrote: Tue Oct 08, 2019 5:42 pm
tadamsmar wrote: Tue Oct 08, 2019 5:15 pm Kindly provide a mathematical proof for that mathematical proposition.
It's not a "mathematical proposition" so much as an empirical observation: the only solution to the Markowitz equations which generates an efficient frontier that contains the L fund allocations is one in which the expected return of the G Fund and F Funds are identical.

If you don't believe me, try it yourself and see what you get.
But the return is not the only factor that determines the allocations. Therefore, your claim is false.
Try to solve the problem: unless you learn to do it for yourself, I suspect you will keep misunderstanding the constraints you’re under with a fund like the G Fund.
"Far more money has been lost by investors preparing for corrections than has been lost in corrections themselves." ~~ Peter Lynch
User avatar
nps
Posts: 1630
Joined: Thu Dec 04, 2014 9:18 am

Re: Improving the TSP [for current participants]

Post by nps »

vineviz wrote: Tue Oct 08, 2019 12:30 pm
tadamsmar wrote: Tue Oct 08, 2019 11:06 am They are probably assuming F has a somewhat higher return than G. Seems to me that is a reasonable assumption and it looks to be consistent with the facts that we know. The allocations depend on the correlation matrices of the 5 TSP assets.
Sorry, but no. The only way to generate an efficient frontier which contains the L fund allocations is to assume that the two funds have the same expected return. If the consultants had assumed that "F has a somewhat higher return than G", the L funds would have much lower allocations to the G Fund than they actually do.
Actually what tadamsmar stated is exactly what they assumed:

https://www.frtib.gov/pdf/minutes/MM-2018Sep-Att5c.pdf

Slide 14 has their expected return assumptions.
User avatar
Topic Author
vineviz
Posts: 14921
Joined: Tue May 15, 2018 1:55 pm
Location: Baltimore, MD

Re: Improving the TSP [for current participants]

Post by vineviz »

nps wrote: Wed Oct 09, 2019 6:13 am
vineviz wrote: Tue Oct 08, 2019 12:30 pm
tadamsmar wrote: Tue Oct 08, 2019 11:06 am They are probably assuming F has a somewhat higher return than G. Seems to me that is a reasonable assumption and it looks to be consistent with the facts that we know. The allocations depend on the correlation matrices of the 5 TSP assets.
Sorry, but no. The only way to generate an efficient frontier which contains the L fund allocations is to assume that the two funds have the same expected return. If the consultants had assumed that "F has a somewhat higher return than G", the L funds would have much lower allocations to the G Fund than they actually do.
Actually what tadamsmar stated is exactly what they assumed:

https://www.frtib.gov/pdf/minutes/MM-2018Sep-Att5c.pdf

Slide 14 has their expected return assumptions.
First, slide 14 shows that my statement was correct, not tadamsmar's: the assumptions in the "previous study" column are the ones used to generate the existing L Fund allocations, and the assumed return for G Fund and F Fund were both 3.5%.

Second, it's true that the current return assumptions (which should be used to construct future L Fund allocations) have modified the expected returns to be more realistic with the expected return of F Fund > G Fund. Both are arguably still indefensible, since the expected return is FAR above the current YTM for each fund and the expected volatility of the G fund is about 50% too high, but leave that aside for now. Despite changing the relative expected returns for the two funds, the proposed relative allocation for the 2018 L Fund allocation hasn't changed at all. Which leads to my third point:

According to the consultant's own assumptions, the L Fund allocations are NOT on their own modeled efficient frontier. Check out slide 15:

Image

Now, in fairness to Aon Hewitt it seems that their mandate is NOT to maximize risk-adjusted return in the classic Sharpe ratio (i.e. efficient frontier) sense but rather to optimize the income replacement ratio for plan participants. Check out slide 18: they are defining "risk" as the 5th percentile outcome for income replacement ratio and "reward" as 50th percentile outcome. This is consistent with their goal as stated in slide 3:
The desired outcome is to create a series of L Funds such that an “average participant” in those L Funds, in combination with the FERS defined benefit plan and Social Security, will be projected to have sufficient assets to maintain a reasonable standard of living throughout retirement.
This is an entirely reasonable goal, even if it is a little opaque without full access to the data they have about plan participants, and it certainly is a quantitative goal. But it means they are NOT trying to allocate the L Funds to be on the Markowitz frontier.

It also means that the L Fund allocation should be appropriate for the average plan participant. Furthermore, in terms of advice I'd recommend the L Funds to every TSP participant before I'd recommend a DIY portfolio. Especially with the more modern, less conservative glide paths.

But virtually NO participant is going to precisely match the AVERAGE participant on every assumed parameter (e.g. age, salary, deferral rate, risk aversion, external assets, life expectancy, etc.). As a result, the L Fund allocations are not likely to be optimal for any investor who is serious about trying a DIY asset allocation. Good enough? Absolutely, but let's not pretend the allocations are magical or even more efficient than they actually are.
"Far more money has been lost by investors preparing for corrections than has been lost in corrections themselves." ~~ Peter Lynch
rkhusky
Posts: 17659
Joined: Thu Aug 18, 2011 8:09 pm

Re: [Thrift Savings Plan - Fund composition and comparisons]

Post by rkhusky »

Wondering if the TSP is expecting interest rates to rise over the next 10 years, and are thus expecting the F Fund returns to be less than the expected yields. And perhaps they were expecting the same thing 10 years ago. Many people have had the same expectations.
User avatar
tadamsmar
Posts: 9972
Joined: Mon May 07, 2007 12:33 pm

Re: Improving the TSP [for current participants]

Post by tadamsmar »

vineviz wrote: Wed Oct 09, 2019 6:57 am
nps wrote: Wed Oct 09, 2019 6:13 am
vineviz wrote: Tue Oct 08, 2019 12:30 pm
tadamsmar wrote: Tue Oct 08, 2019 11:06 am They are probably assuming F has a somewhat higher return than G. Seems to me that is a reasonable assumption and it looks to be consistent with the facts that we know. The allocations depend on the correlation matrices of the 5 TSP assets.
Sorry, but no. The only way to generate an efficient frontier which contains the L fund allocations is to assume that the two funds have the same expected return. If the consultants had assumed that "F has a somewhat higher return than G", the L funds would have much lower allocations to the G Fund than they actually do.
Actually what tadamsmar stated is exactly what they assumed:

https://www.frtib.gov/pdf/minutes/MM-2018Sep-Att5c.pdf

Slide 14 has their expected return assumptions.
First, slide 14 shows that my statement was correct, not tadamsmar's: the assumptions in the "previous study" column are the ones used to generate the existing L Fund allocations, and the assumed return for G Fund and F Fund were both 3.5%.

Second, it's true that the current return assumptions (which should be used to construct future L Fund allocations) have modified the expected returns to be more realistic with the expected return of F Fund > G Fund. Both are arguably still indefensible, since the expected return is FAR above the current YTM for each fund and the expected volatility of the G fund is about 50% too high, but leave that aside for now. Despite changing the relative expected returns for the two funds, the proposed relative allocation for the 2018 L Fund allocation hasn't changed at all. Which leads to my third point:

According to the consultant's own assumptions, the L Fund allocations are NOT on their own modeled efficient frontier. Check out slide 15:

Image

Now, in fairness to Aon Hewitt it seems that their mandate is NOT to maximize risk-adjusted return in the classic Sharpe ratio (i.e. efficient frontier) sense but rather to optimize the income replacement ratio for plan participants. Check out slide 18: they are defining "risk" as the 5th percentile outcome for income replacement ratio and "reward" as 50th percentile outcome. This is consistent with their goal as stated in slide 3:
The desired outcome is to create a series of L Funds such that an “average participant” in those L Funds, in combination with the FERS defined benefit plan and Social Security, will be projected to have sufficient assets to maintain a reasonable standard of living throughout retirement.
This is an entirely reasonable goal, even if it is a little opaque without full access to the data they have about plan participants, and it certainly is a quantitative goal. But it means they are NOT trying to allocate the L Funds to be on the Markowitz frontier.

It also means that the L Fund allocation should be appropriate for the average plan participant. Furthermore, in terms of advice I'd recommend the L Funds to every TSP participant before I'd recommend a DIY portfolio. Especially with the more modern, less conservative glide paths.

But virtually NO participant is going to precisely match the AVERAGE participant on every assumed parameter (e.g. age, salary, deferral rate, risk aversion, external assets, life expectancy, etc.). As a result, the L Fund allocations are not likely to be optimal for any investor who is serious about trying a DIY asset allocation. Good enough? Absolutely, but let's not pretend the allocations are magical or even more efficient than they actually are.
I am not sure you are interpreting that plot correctly. It could be that the efficient frontier line is based on their new modeling as if they just jump immediately to the new L Fund allocations instead of using a 10-year glide path, whereas the lower performance point estimates for the various L Funds are based on what that actual gradually changing glide path allocations are.

You do seem to have a point on the previous study geometric return. But, it depends on what the meaning of "return" is. I think that the annual return of F is projected to be higher than G but volatility drag causes them to have the same geometric return over the 10-year period.

You can look at the glide path for the L funds here, using a slider. This page reflects the Hewitt glide path from 20% to 30% equities for the L Income Fund:

https://www.tsp.gov/InvestmentFunds/Fun ... ncome.html

This reveals an interesting glide path side effect. A retired fed with his/her nest egg in the L Income fund is going to have a risk increase over the next 10 years of their retirement, contrary to the usual recommended asset allocation schedule. (Mostly harmless, I suppose.)

Anyway, it does not seem to me that the L Fund G vs F allocations change all that much in this new analysis.
HeelaMonster
Posts: 1038
Joined: Sat Aug 10, 2019 11:46 am

Re: Improving the TSP [for current participants]

Post by HeelaMonster »

tadamsmar wrote: Wed Oct 09, 2019 9:35 am You can look at the glide path for the L funds here, using a slider. This page reflects the Hewitt glide path from 20% to 30% equities for the L Income Fund:

https://www.tsp.gov/InvestmentFunds/Fun ... ncome.html

This reveals an interesting glide path side effect. A retired fed with his/her nest egg in the L Income fund is going to have a risk increase over the next 10 years of their retirement, contrary to the usual recommended asset allocation schedule. (Mostly harmless, I suppose.)
This is of relevance to me, as a newly retired fed with all TSP assets "dialing down" to the L-Income fund. I had noticed that the allocation to G fund was slated to decrease from current 73% to 65% over the next 10 years, which seemed counterintuitive. That doesn't bother me from the risk perspective (and as pointed out in this relocated thread, G fund introduces risk of another kind)..... but I haven't understood why this gradual shift is scheduled to occur, once one arrives at their "final resting place" in the Income Fund(?).
User avatar
Topic Author
vineviz
Posts: 14921
Joined: Tue May 15, 2018 1:55 pm
Location: Baltimore, MD

Re: Improving the TSP [for current participants]

Post by vineviz »

tadamsmar wrote: Wed Oct 09, 2019 9:35 am I am not sure you are interpreting that plot correctly. It could be that the efficient frontier line is based on their new modeling as if they just jump immediately to the new L Fund allocations instead of using a 10-year glide path, whereas the lower performance point estimates for the various L Funds are based on what that actual gradually changing glide path allocations are.
The line shows the efficient frontier based on their new assumptions about expected returns, expected volatility, and expected correlations (as shown in slide 14).

The points show the location of the proposed 2018 L fund allocations (as shown on slide 25).

tadamsmar wrote: Wed Oct 09, 2019 9:35 am I think that the annual return of F is projected to be higher than G but volatility drag causes them to have the same geometric return over the 10-year period.
That doesn't work because neither fund has enough expected volatility to make a difference. Plus the directionality is wrong: with the same expected geometric returns, the higher volatility of the F Fund would lead to bigger allocations in the efficient frontier not lower ones. That's why the G Fund was never allocated all the bond sleeve even when the returns were assumed to be equal.
"Far more money has been lost by investors preparing for corrections than has been lost in corrections themselves." ~~ Peter Lynch
User avatar
nps
Posts: 1630
Joined: Thu Dec 04, 2014 9:18 am

Re: Improving the TSP [for current participants]

Post by nps »

vineviz wrote: Wed Oct 09, 2019 6:57 am
nps wrote: Wed Oct 09, 2019 6:13 am
vineviz wrote: Tue Oct 08, 2019 12:30 pm
tadamsmar wrote: Tue Oct 08, 2019 11:06 am They are probably assuming F has a somewhat higher return than G. Seems to me that is a reasonable assumption and it looks to be consistent with the facts that we know. The allocations depend on the correlation matrices of the 5 TSP assets.
Sorry, but no. The only way to generate an efficient frontier which contains the L fund allocations is to assume that the two funds have the same expected return. If the consultants had assumed that "F has a somewhat higher return than G", the L funds would have much lower allocations to the G Fund than they actually do.
Actually what tadamsmar stated is exactly what they assumed:

https://www.frtib.gov/pdf/minutes/MM-2018Sep-Att5c.pdf

Slide 14 has their expected return assumptions.
First, slide 14 shows that my statement was correct, not tadamsmar's: the assumptions in the "previous study" column are the ones used to generate the existing L Fund allocations, and the assumed return for G Fund and F Fund were both 3.5%.
I'm not sure that Aon's previous study is the same as what is currently implemented. More to the point, when the assumed return is different, the allocation recommendation for G and F doesn't change at all (except for L income).
User avatar
tadamsmar
Posts: 9972
Joined: Mon May 07, 2007 12:33 pm

Re: Improving the TSP [for current participants]

Post by tadamsmar »

vineviz wrote: Wed Oct 09, 2019 11:48 am
tadamsmar wrote: Wed Oct 09, 2019 9:35 am I think that the annual return of F is projected to be higher than G but volatility drag causes them to have the same geometric return over the 10-year period.
That doesn't work because neither fund has enough expected volatility to make a difference.
The sd of F is 5.3% and the sd of G is 1.2%. That is big enough to make a difference.
Plus the directionality is wrong: with the same expected geometric returns, the higher volatility of the F Fund would lead to bigger allocations in the efficient frontier not lower ones.
The directionality is right since F has the high SD so it has the higher drag, see here:

https://en.wikipedia.org/wiki/Volatility_tax

The equation is geometric mean = arithmetic mean - sd2/2 , see here:

https://www.kitces.com/blog/volatility- ... t-returns/

If F and G have the same annual geometric return then F has the higher average annual arithmetic return, because the arithmetic return has to get drug down more by the higher sd of F.

We are talking about the average annual returns of the individual assets in isolation from the other assets. Allocations are not a factor in this analysis, that has to do with how the 5 assets play together.
That's why the G Fund was never allocated all the bond sleeve even when the returns were assumed to be equal.
I'd say that would happen farther right on the efficient frontier than the location of the L Income Fund. You just don't see it because they never show it.
User avatar
Topic Author
vineviz
Posts: 14921
Joined: Tue May 15, 2018 1:55 pm
Location: Baltimore, MD

Re: Improving the TSP [for current participants]

Post by vineviz »

tadamsmar wrote: Wed Oct 09, 2019 7:51 pm The sd of F is 5.3% and the sd of G is 1.2%. That is big enough to make a difference.
First off, those volatility numbers are much higher than they actually are. Even if they weren't , they are still too small to have a material impact on the allocations. Please, do the math on this if think you're correct. You'll see that you're not, and I know that because I have already done the math myself. I'm not just making stuff up here: I do this for a living.
tadamsmar wrote: Wed Oct 09, 2019 7:51 pmPlus the directionality is wrong: with the same expected geometric returns, the higher volatility of the F Fund would lead to bigger allocations in the efficient frontier not lower ones.
The directionality is right since F has the high SD so it has the higher drag, see here:[/quote]

You can't just look something up an Wikipedia and become an expert on this. You're completely off base on the quantitative stuff and politically I don't even understand what you're trying to defend here. The fact that Aon itself produced an efficient frontier diagram which shows that the L Funds are NOT on it should settle the original question, right?
"Far more money has been lost by investors preparing for corrections than has been lost in corrections themselves." ~~ Peter Lynch
User avatar
tadamsmar
Posts: 9972
Joined: Mon May 07, 2007 12:33 pm

Re: Improving the TSP [for current participants]

Post by tadamsmar »

vineviz wrote: Wed Oct 09, 2019 8:12 pm
tadamsmar wrote: Wed Oct 09, 2019 7:51 pm The sd of F is 5.3% and the sd of G is 1.2%. That is big enough to make a difference.
First off, those volatility numbers are much higher than they actually are. Even if they weren't , they are still too small to have a material impact on the allocations. Please, do the math on this if think you're correct. You'll see that you're not, and I know that because I have already done the math myself. I'm not just making stuff up here: I do this for a living.
I did the math. The average annual arithmetic mean of F is 3.8. The average annual arithmetic mean of G is 3.5.
The fact that Aon itself produced an efficient frontier diagram which shows that the L Funds are NOT on it should settle the original question, right?
Those L Funds are on a different efficient frontier calculated with different inputs. Didn't you say that in one of your posts?
rkhusky
Posts: 17659
Joined: Thu Aug 18, 2011 8:09 pm

Re: [Thrift Savings Plan - Fund composition and comparisons]

Post by rkhusky »

Perhaps the TSP is thinking about the likelihood of time periods such as this when determining G Fund allocations:
Image
User avatar
tadamsmar
Posts: 9972
Joined: Mon May 07, 2007 12:33 pm

Re: Improving the TSP [for current participants]

Post by tadamsmar »

HeelaMonster wrote: Wed Oct 09, 2019 10:26 am
tadamsmar wrote: Wed Oct 09, 2019 9:35 am You can look at the glide path for the L funds here, using a slider. This page reflects the Hewitt glide path from 20% to 30% equities for the L Income Fund:

https://www.tsp.gov/InvestmentFunds/Fun ... ncome.html

This reveals an interesting glide path side effect. A retired fed with his/her nest egg in the L Income fund is going to have a risk increase over the next 10 years of their retirement, contrary to the usual recommended asset allocation schedule. (Mostly harmless, I suppose.)
This is of relevance to me, as a newly retired fed with all TSP assets "dialing down" to the L-Income fund. I had noticed that the allocation to G fund was slated to decrease from current 73% to 65% over the next 10 years, which seemed counterintuitive. That doesn't bother me from the risk perspective (and as pointed out in this relocated thread, G fund introduces risk of another kind)..... but I haven't understood why this gradual shift is scheduled to occur, once one arrives at their "final resting place" in the Income Fund(?).
I guess you could use a allocation to F2020 and f2030 to get your current equity allocation to around 30%. I think that would nullify the counterintuitive allocation schedule.

It's just a side-effect of the 10-year glide path. The glide path is there so they don't bump then L Income fund from 20% to 30% equities immediately and they don't have abrupt changes to the allocations of other L funds.
User avatar
Topic Author
vineviz
Posts: 14921
Joined: Tue May 15, 2018 1:55 pm
Location: Baltimore, MD

Re: Improving the TSP [for current participants]

Post by vineviz »

tadamsmar wrote: Thu Oct 10, 2019 6:40 am I did the math. The average annual arithmetic mean of F is 3.8. The average annual arithmetic mean of G is 3.5.
First, those numbers don't address the flaw in your earlier argument.

The relevant question is this: are the geometric and arithmetic means of the two funds different enough to produce a significantly different efficient frontier allocation?

So, for a start you should be comparing the arithmetic and geometric mean of each fund. Not the arithmetic mean of each fund to the other. I don't know if you were looking at a historical period you used or computing hypothetic returns using the consultant assumptions, but I just ran the numbers for 9/12 to 9/19 (a period in which the yield of the 10-year Treasury ended pretty much where it started, so no net increase or decrease in rates).

For the F Fund, the arithmetic mean return was 3.18% and the geometric mean return was 3.14%. For the G Fund, the arithmetic mean return was 2.20% and the geometric mean return was 2.20%. The so-called "volatility drag" of the F Fund was 0.04%, which is miniscule and certainly not enough to have a material impact on the allocations you'd derive for the Markowitz frontier.
"Far more money has been lost by investors preparing for corrections than has been lost in corrections themselves." ~~ Peter Lynch
User avatar
tadamsmar
Posts: 9972
Joined: Mon May 07, 2007 12:33 pm

Re: Improving the TSP [for current participants]

Post by tadamsmar »

vineviz wrote: Thu Oct 10, 2019 8:59 am
tadamsmar wrote: Thu Oct 10, 2019 6:40 am I did the math. The average annual arithmetic mean of F is 3.8. The average annual arithmetic mean of G is 3.5.
First, those numbers don't address the flaw in your earlier argument.

The relevant question is this: are the geometric and arithmetic means of the two funds different enough to produce a significantly different efficient frontier allocation?

So, for a start you should be comparing the arithmetic and geometric mean of each fund. Not the arithmetic mean of each fund to the other. I don't know if you were looking at a historical period you used or computing hypothetic returns using the consultant assumptions, but I just ran the numbers for 9/12 to 9/19 (a period in which the yield of the 10-year Treasury ended pretty much where it started, so no net increase or decrease in rates).

For the F Fund, the arithmetic mean return was 3.18% and the geometric mean return was 3.14%. For the G Fund, the arithmetic mean return was 2.20% and the geometric mean return was 2.20%. The so-called "volatility drag" of the F Fund was 0.04%, which is miniscule and certainly not enough to have a material impact on the allocations you'd derive for the Markowitz frontier.
My earlier argument was that this is not true:
vineviz wrote: Tue Oct 08, 2019 12:30 pm Sorry, but no. The only way to generate an efficient frontier which contains the L fund allocations is to assume that the two funds have the same expected return. If the consultants had assumed that "F has a somewhat higher return than G", the L funds would have much lower allocations to the G Fund than they actually do.
(But it is perhaps true if you mean geometric return.)

Nothing more. I did not say it was important.

Heck, earlier I intimated that the whole darn efficient frontier fuzzy math calculation was arguably unimportant.

I kind of have a fondness for the idea that misinformation posted on Bogleheads does not go undisputed, like the earlier claim on this thread that the L Fund allocations were determined by the relative popularity of the five available assets.
User avatar
Topic Author
vineviz
Posts: 14921
Joined: Tue May 15, 2018 1:55 pm
Location: Baltimore, MD

Re: Improving the TSP [for current participants]

Post by vineviz »

tadamsmar wrote: Thu Oct 10, 2019 9:37 am My earlier argument was that this is not true:
vineviz wrote: Tue Oct 08, 2019 12:30 pm Sorry, but no. The only way to generate an efficient frontier which contains the L fund allocations is to assume that the two funds have the same expected return. If the consultants had assumed that "F has a somewhat higher return than G", the L funds would have much lower allocations to the G Fund than they actually do.
Your argument is no stronger now then it was then: there is, in fact, no way to generate an efficient frontier which contains the L fund allocations without assuming the two funds have the same expected return.

We've now seen the TSP consultants literally verify this: their own presentation shows that the former assumption for expected return was exactly what I said it had to be (i.e. the same) and on the very next page shows how far below the efficient frontier the funds are when that assumption is abandoned.

Seriously, the consultants confirmed that what I said was correct so I don't understand the urge to flail around for new ways to assert otherwise.
"Far more money has been lost by investors preparing for corrections than has been lost in corrections themselves." ~~ Peter Lynch
Post Reply