Permanent Portfolio May 6 2010 Reaction

[craigr]

I don't watch the markets each day, but yesterday was just too interesting to not take a couple looks. The stocks corrected before LT bonds could really take off, but if it had continued it might have been 2008 all over again. Just for grins I took a snapshot of the past five days which have been very volatile. The LT bonds are moving like a mirror to to the stocks (the spike down could have been met with a spike up if things in the stock market did not correct):

TLT Treasury LT ETF in Blue. Vanguard Total Stock Market in Red

[fnord123]

I actually did make a change to my PP based on yesterday's market action.  Specifically, I calculated the price at which my stock holdings would be 20% of my portfolio (they are 21.29% currently).  Then I entered a GTC limit order to buy enough SPY @ that price to bring the holdings back to 25%.  Basically I entered an auto-rebalance order in my broker that will take advantage of similar such volatility in the future.

I also calculated similar thresholds for TLT and GLD, but since both of them are much further from rebalancing (24.36% and 25.92% respectively), I didn't bother entering an order.  Maybe if they get closer to the 20% guard band I might...

[Roy]

What's quite interesting is that if you consider foreign currency as an alternative to gold, and that a diverse range of stocks can reduce the need for a CASH damper, then reducing PP's gold down from 25% to 15% and reducing CASH down to 10%, leaving LT's as-is at 25%, but then spreading across domestic and foreign stocks (so some foreign currency type exposure to counter the reduced amount of gold) can vastly improve returns.

From Simba's spreadsheet (1972 to 2009), a blend of 10% cash (Money Market), 15% gold, 25% LT's and 10% each in Large Cap Value, Small Cap Value, Small Cap Growth, Emerging Markets and International Value (the last two acting as a form of foreign currency holding to counter-balance holding less gold) produced a annualised gain that was 2.5% more than the classic PP.  Beating the Coffee House by 1% annualised and with lower standard deviation (risk), whilst retaining many of the PP like qualities.  Switching the MM for ST produced a slightly higher annualised of 11.78% for around the same standard deviation (10.52 vs 10.55). 

Better than the PP's 9.1% annualised, 8.02% standard deviation and up until 2008 the drawdowns were much the same as PP's.  In 2008 however the drawdown did rise to -13%.  So there is more risk involved (risk/rewards generally go hand-in-hand).

I have found that it is easy to press buttons dispassionately on some backtest spreadsheet over the morning coffee.

Many posters from various forums have "discovered" all sorts of backtested tinkerings that were "better" than the 4x25 PP, or "improvements" to it, or indeed, better than anything else.  Being based wholly on hindsight, the spreadsheet backtest game avoids any of the emotional challenges that may have threatened those investors who held those hypothetical cobblings through in-the-moment rumors, panics, valuations arguments, and rival spreadsheet backtest musings du jour.

There is always some higher-returning portfolio with better Standard Deviation that wins the spreadsheet game.  The problem is this game is like an ivory tower, military theory wargame compared with "seeing the elephant" that is real combat, and likely far less useful than even that.  The spreadsheet doesn't have a button for individual investor emotion.

When sizable downturns occur (like 2008), the one thing few do is calmly consider the gains on "either side," assuming they had been consistently invested in the first place, or could have been (as some of the constructs are hypothetical synthetics).  The forward side can't be known and the recent past is forgotten in the present terror;  it's happening right now, and if the DOW drops another thousand, we'll really see some forgetfulness on the joys of '09.  What is done when things get really bad, however, are things like "Plan B" confessions, re-assessments of actual risk profile, strategy shifts to things like Hussman funds, et al, or outright capitulations.  Those all interest me far more because passion is involved.  Emotion in investing is truth I can understand.

But being an excitable boy, that's just the way I tend to see these things.

Roy

[macclary]

Hi Roy, I agree that people need to factor themselves into their backtesting. "Are you going to get scared away from your plan, or are you going to get board of it?"

I would hesitate to be so dismissive of backtesting and optimizing. Harry Browne and his colleague did a great deal of that type of work when creating the PP and preceding portfolio recommendations. I think backtesting and portfolio optimizing is more than a game.

That said though some faulty optimization strategies can result in a worse portfolio going forward.  Try walk forward testing for mean-variance optimization of high return/volatility type portfolios and I think you will see pro-cyclical and overall worse performance than an un-optimized portfolio.

Better optimization strategies can result in a portfolio with improved out-of sample performance. I think RiskCog falls in this category.

Additionally there are some optimization approaches that don't reliable improve or worsen a portfolio. Maybe slicing and dicing sectors based on 3 year sharpe ratios would fall here...

[Roy]

I would hesitate to be so dismissive of backtesting and optimizing. Harry Browne and his colleague did a great deal of that type of work when creating the PP and preceding portfolio recommendations. I think backtesting and portfolio optimizing is more than a game.

Hi, Macclary,

In a broad sense, I do think we need to understand how risk premia have worked, and the past helps some in demonstrating that. 

Correlations of asset classes are useful—to a point (as you suggest, especially with various S/D approaches, and where movements can become highly correlated and at bad times).  Beyond that, and from what I've seen posted, the "tools" provide little that is useful and indeed get carried to the point of gaming.  There are degrees to everything.  But beyond entertainment, I see limits to their usefulness.

Some of these "optimal or best" allocations could never really have existed (certainly not as index-like approaches, or they use multiple funds or benchmarks across time, or due to being synthetic) and it is doubtful many would actually stick to the weird percentages ascribed to various "optimal" portfolios.  There are endless iterations of these.  (Most of that sort appeared on other boards.)  Also, Standard Deviation is just one kind of risk but is ballyhooed as being far more than that (on many professional websites too).  So, I see lots of confounding variables to consider.

I have no doubt Harry considered a number of approaches;  I imagine 4x25 was not his first guess.  I would be interested to know the details of what he did in this manner.  Can you point me to some specifics that illustrate the various paths he may have considered before settling as he did?

Roy

[craigr]

I have no doubt Harry considered a number of approaches;  I imagine 4x25 was not his first guess.  I would be interested to know the details of what he did in this manner.  Can you point me to some specifics that illustrate the various paths he may have considered before settling as he did?

From my discussions with his former business colleague and publisher they did actually pay quite a bit of money in the 1970's to have some computer analysis performed on portfolio designs. But this was also combined with a study of history and understanding of Austrian economics to work out the details.

Backtesting, as Browne pointed out as well, can really only be used to disprove ideas. It can't really be used to prove anything. We can use it to show that some asset allocation didn't work in the past. But we really can never state that some other allocation is going to work forever going forward. We can only use backtesting to throw out models that haven't worked in a way we wanted so we don't spend any more time considering them.

This is the main problem I have with stock/bond only portfolios. They have failed numerous times in the past to either provide real after inflation returns for extended periods (10+ years) or have had catastrophic drops in value. The backtests show this conclusively so there was no reason why 2008 or even 2000-2010 should have been a surprise to anyone. So in my world-view they are a broken strategy and we need to look somewhere else. Backtesting showed this as a possibility before it happened (again). The PP has not experienced these events in the past which just shows that it is a contender but doesn't really mean it's guaranteed to always work going forward. Although I will say in its defense that it's survived some pretty bad markets compared to competitors and did OK.

So backtesting can be useful, but it's also a dangerous tool if used in ways not intended. I think a lot of investing books use it in ways not intended for instance and this leads to bad outcomes.

[Roy]

I have no doubt Harry considered a number of approaches;  I imagine 4x25 was not his first guess.  I would be interested to know the details of what he did in this manner.  Can you point me to some specifics that illustrate the various paths he may have considered before settling as he did?

From my discussions with his former business colleague and publisher they did actually pay quite a bit of money in the 1970's to have some computer analysis performed on portfolio designs. But this was also combined with a study of history and understanding of Austrian economics to work out the details.

Backtesting, as Browne pointed out as well, can really only be used to disprove ideas. It can't really be used to prove anything. We can use it to show that some asset allocation didn't work in the past. But we really can never state that some other allocation is going to work forever going forward. We can only use backtesting to throw out models that haven't worked in a way we wanted so we don't spend any more time considering them. 

So backtesting can be useful, but it's also a dangerous tool if used in ways not intended. I think a lot of investing books use it in ways not intended for instance and this leads to bad outcomes.

Funny, as Craigr, suggests I have used it to show how inexpensive index funds have beaten or matched the record of "famous" low-cost active funds using equivalent percentages of stocks and bonds.  The point being that style-drifting and stock selection was not as comforting and productive as many believed—even with certain inexpensive and proud active funds.  The active management (which, by necessity, did cost a bit more) resulted in style-drifting, that was the likely cause of underperformance. This came as a surprise or invoked sharp defense on the part of the active fund supporters.  And the belief in those active funds continued despite repeated displays of this simple data.  Which, of course, makes sense in a world where emotions, shamans, and mystical beliefs, not spreadsheet buttons, hold sway.


It is curious that after his work, using then-complex computer models, that Brown arrived at the wonderfully symmetrical, boringly plain 4x25 (of course, he lacked the better spreadsheet toys we have available now!).  Now there probably is a better way than 4x25, and we'll know that way in another 40 years.

I wonder sometimes if Browne's formula is so plain that this itself invokes doubt.  I mean, if it had 10 asset classes sliced into wildly-fractional allocation percentages, it would then surely suggest a sophistication that just had to be efficient, or optimal, or best, or whatever that buzzword is I'm looking for.  Yet there the PP sits in all its quadrangled plainness.  It's to laugh...

Again, I do think that in a general sense, the deep past informs us on correlations (especially the sharpest differentials) and demonstrates the economic principles of risk/return on which the basic premia, and common sense, rest.  But most of the backtesting I see is done with more than a hint of "sheer proof" of how that particular cobbling might be so going forward, and the ease and dispassion of the button pushing creates an UNreality, and the things that make it unreal can not be represented with even hindsight data.  It can entertain, though.

Roy