Fat Tail Minimisation compared to 4x25 Permanent Portfolio

[Clive]

Tweaking Simba's spreadsheet to replace yearly inflation values with yearly TIP gain values and then comparing the 'real' gains (after adjusting for TIP) for both the Fat Tail Minimisation (FTM) and the Permanent Portfolio (PP)



it would appear that the FTM was much more consistent at acting like a TIP+ type investment.

[steve r]

I am curious why you would replace actual inflation data with TIP values given that for most the years were simulated TIPS returns ?

[Clive]

why you would replace actual inflation data with TIP values

To provide a more direct graphical comparison

Similar real gain progressions can equally be seen when substituting inflation with T-Bill or Short Term Treasury or 5 year Treasury or 5 year treasury ladder or STT/LTT barbell ...etc. i.e. actuals (non-synthetic).

Unlike Fat Tail Minimisation, it would seem the Permanent Portfolio - a claimed safe investment, is generally neither consistently safe nor as rewarding as other choices of 'safe' investments.

[happymob]

Is the 35% TIP allocation based on buying 30-year TIPS? If so, I have a couple questions:

1) How did this account for the period where 30-year TIPS weren't auctioned? Just replace them with 20-year TIPS?

2) How does one backtest to 1972 when TIPS weren't offered until 1997? How is the earlier 25 years (1972- 1997) data for FTM meaningful?

Much like Permanent Portfolio folks starting at 1972 (the date obviously chosen tdue to gold restrictions being removed), I guess I don't see how backtesting this to 1972 makes any more sense. It also would seem to benefit more from the great bull market in bonds (in this case TIPS) then the PP due to higher allocation (35% in LT TIPS vs. 25% in LT treasuries). But I suspect that is simply cherry-picking after the fact. I guess I simply don't know quite what to make of this chart.

[dkturner]

Is the 35% TIP allocation based on buying 30-year TIPS? If so, I have a couple questions:

1) How did this account for the period where 30-year TIPS weren't auctioned? Just replace them with 20-year TIPS?

2) How does one backtest to 1972 when TIPS weren't offered until 1997? How is the earlier 25 years (1972- 1997) data for FTM meaningful?

Much like Permanent Portfolio folks starting at 1972 (the date obviously chosen tdue to gold restrictions being removed), I guess I don't see how backtesting this to 1972 makes any more sense. It also would seem to benefit more from the great bull market in bonds (in this case TIPS) then the PP due to higher allocation (35% in LT TIPS vs. 25% in LT treasuries). But I suspect that is simply cherry-picking after the fact. I guess I simply don't know quite what to make of this chart.

The modern personal computer, an inquisitive mind, and a lot of free time is a potent combination.

[Call_Me_Op]

I question how much we can rely on back-testing - even over a 40 year period.

I will also point out that this "fat-tail minimization" portfolio did not hold up nearly as well as PP in 2008, FWIW.

Thirdly, using yearly TIPS gains (simulated or not) as a baseline seems rather arbitrary.

Fourthly, since we are relying on back-testing, why not replace all bonds with 5 year treasuries, which beat the pants off of the bond combination you chose?

[steve r]

why you would replace actual inflation data with TIP values

To provide a more direct graphical comparison

Similar real gain progressions can equally be seen when substituting inflation with T-Bill or Short Term Treasury or 5 year Treasury or 5 year treasury ladder or STT/LTT barbell ...etc. i.e. actuals (non-synthetic).

Well - call me skeptical of the "technique." I guess that is what makes a forum. Are we not concerned with real gain over inflation and not returns of an asset that is a large chunck of the portfolio.

I wonder what the graph would look like if we called real gain progression by substituting inflation with the returns from gold. I am not sure why we would do this but I suspect someone would note gold is expected to have zero real return (i.e. inflation). I could then claim that PP is more consistent with a GOLD+type investment. I would be skeptical of this technique as well.

Nit picking comment I know - I suspect a graph getting real returns with actual inflation would be roughly similar with differences somewhat muted and smaller differences in Sortino Ratio.

Bottom line, over the period, CAGR is higher with FTM (by a full percentage point), but Sortino is higher with PP in nominal terms.

Several critiques of both models. As various assets become easier to invest in 1) Corrleations are likely to be larger 2) Return over TSM are likely to be smaller (gold, Emerging Markets, SCV).

If you knew the highest two returning asset classes over the next 40 years and could only put 30% in them in total ... that would probably be the way to go.

[Clive]

Using yearly TIPS gains (simulated or not) as a baseline seems rather arbitrary

Using etfreplay.com backtest/ETF Portfolio option to plot individual years for 50-50 SSO/TIP, benchmarked to SPY, indicates that for all years 2007 onwards the 50-50 relatively outperformed SPY. Conceptually if an asset allocation consistently relatively outperforms TIP then the potential is even greater. Generally the PP failed that criteria whilst FTM was much more viable. Replacing TIP with FTM provided a 1% annualised higher reward for 50-50 SSO/FTM compared to SPY across 2007 to YTD 2012 when yearly rebalanced.

[steve r]

Using yearly TIPS gains (simulated or not) as a baseline seems rather arbitrary

Using etfreplay.com backtest/ETF Portfolio option to plot individual years for 50-50 SSO/TIP, benchmarked to SPY, indicates that for all years 2007 onwards the 50-50 relatively outperformed SPY. Conceptually if an asset allocation consistently relatively outperforms TIP then the potential is even greater. Generally the PP failed that criteria whilst FTM was much more viable. Replacing TIP with FTM provided a 1% annualised higher reward for 50-50 SSO/FTM compared to SPY across 2007 to YTD 2012 when yearly rebalanced.

Wow!
Not sure I will be buying a levered ETF anytime soon (another discussion) but wow - even in 2008.

edit: actually I am showing the SSO/TIP underperformed in 09, 10, and YTD so I am not sure what you mean "for all years"... still less voltility with comprable returns.

[steve r]

I think I get what you are saying ... PP does not hold up well in "real" terms (leaving aside how to measure real) and if ways to outperform TIPS can be found, might as well use them (in lieu of TIPS).

Pluging in things in the Simba data, I was shocked how poorly PP does after inflation. Sortino (my favorite measure) fell from 1.78 to 0.02 ... The ratio turns negative if you start in 1975

Playing around here ... plug in Gold for Tips in FTM ... lead to some impressive results (Sortino at 0.50 - a great number after inflation) ... like I said, playing around here I cannot imagine ever being 35% gold.

Learned something today ... thanks.

[KyleAAA]

Correct me if I'm wrong, but TIPS don't actually track inflation all that closely in the short-term. Correct?

[Clive]

actually I am showing the SSO/TIP underperformed in 09, 10, and YTD

I suspect that's because you're looking at the yearly figures of buy-and-hold multiple years - look at each year individually instead - i.e. as though you'd rebalanced back to 50-50 SSO/TIP at the start of each new year (select 2009 for instance in the top right 'Period' dropdown and I see 28.1% for the 50-50 versus 26.4% for the SPY, 16.5%/15.1% for 2010, 5.2%/1.9% 2011, 19.2%/16.3% YTD).

[Call_Me_Op]

That's correct. The market value of TIPS does not track inflation with any degree of exactitude. In fact, it may be essentially uncorrelated.

[Clive]

... like I said, playing around here I cannot imagine ever being 35% gold

etfreplay 33.3% UGLD, 66.7% TIP benchmarked to GLD (limited history - 50-50 UGL/TIP has longer history).

Once yearly rebalanced and 36% gold exposure via 12% UGLD, rest in TIP's has a max single year loss risk of -12% assuming TIPS to be risk-free.

[athrone]

Looks to me like you are just plotting the difference between TSM and SCV/EM.