Gyroscopic Investing

Posted: **Wed Mar 13, 2013 12:43 am**

In 1996, Bridgewater Associates established the All Weather principles for asset allocation, which have now been more broadly adopted under the banner “Risk Parity.”? In 2004, Mr. Dalio wrote an article in which he explained these principles. That article is reprinted here, with relevant updates.

http://www.docyoushare.com/file/index.php?f=eil1MHIu

Posted: **Thu Mar 14, 2013 10:07 pm**

1996, didn't the other article in the other thread say they had been discussing it in the 1970's? Oh well, no matter.

I'm deeply skeptical of attempting to so finely tune future risk and performance based on past risk and performance.

Calling it "engineering" is in vogue, but in reality the only relationship to engineering is math. Engineering is based on testable, falsifiable and documented principles. Financial engineering of this sort is based on theory and assumptions resulting in the famous GIGO principle with no way of knowing in advance.

MPT is based on academic theory with innumerable assumptions to try and make it applicable to the real world.

For example, risk. In MPT risk has meaning. In the real world future risk is unknowable and past risk is an impossible concept so when trying to apply MPT to the real world they equate risk and beta. Beta can be calculated from historical data. But beta isn't risk and equating it invalidates in the real world any plausibility MPT might have in the theoretical.

Posted: **Thu Mar 14, 2013 10:21 pm**

Well, I don't think his strategies are overly engineered really. He really just offers something very similar to the PP, except with different levels of leverage applied so that the investor can take risk commensurate with their risk tolerance.

FWIW I have found that for the PP you can dial in the amount of leverage you take with the PP by adjusting the bond duration (duration is very similar to leverage for bonds): http://gyroscopicinvesting.com/forum/ht ... 148#p50148

I like Dalio and I think studying his fund and strategy has helped me learn more about the PP and macro based investing than any other source (except for this forum and HB himself of course).

Posted: **Thu Mar 14, 2013 11:39 pm**

Ag, I thiink you're confusing the two definitions of beta. It means both the traditional correlated volatility to an index that is standard in CAPM as well as the more recent definition for exposure to any asset's risk. I do not know why they conflated the two definitions in one term. Typical "ivory tower" academics, probably.

There's nothing engineered about the AWP other than they use leveraging and deleveraging to normalize the risk across all the asset classes relative to each other. That is the single crucial difference between AWP and PP and it solves a major flaw of the PP being overexposed to gold's beta (in the risk exposure sense) and "tight money".

The problem is I'm not quite convinced us mere mortals have the ability to delevarage gold, stocks and bonds and leverage up cash so all can meet in the sweet spot. Perhaps melveyr's duration concept is the answer? But I worry about not having cash as even the AWP has cash. A middle compromise from the barbell just doesn't seem like it will work.

Posted: **Thu Mar 14, 2013 11:56 pm**

MG,

I have noticed your posts moving towards more quanty type stuff which I like

You might be interested in an extremely interesting formula which I stumbled upon. Most people assemble portfolios based on "dollar" allocations. However, you can also look at a portfolios "risk allocation." To do this for each asset you look at the weighted beta of each asset in the portfolio. In this context, the individual assets beta is with respect to the total portfolio (normally people calculate beta with respect to the stock market). So for example to find the risk allocation of the PP you would do the following.

Gold Risk: (weight of gold in portfolio) * (cov(gold returns, portfolio returns) / portfolio variance)
Stock Risk: (weight of stocks in portfolio) * (cov(stock returns, portfolio returns) / portfolio variance)
LTT Risk: (weight of LTT in portfolio) * (cov(LTT returns, portfolio returns) / portfolio variance)

A 3 way split between gold, LTT, and stocks has historically actually had unequal risk allocation...

Risk Allocation
Stocks: 30.1%
LTT: 24.8%
Gold: 45%

So in terms of risk allocation, the PP has historically been overweight gold even though the dollar amounts make it appear neutral.

For the 1978-2011 period the following dollar weightings were what provided equal risk contribution between the assets.

Stocks: 34%
LTT: 37%
Gold: 28%

In some sense this is reassuring that it is not that far off from the traditional PP! However, it helps make an argument that the PP is slightly overweight gold if you are aiming for neutrality between the assets.

Posted: **Fri Mar 15, 2013 12:16 am**

melveyr wrote: I have noticed your posts moving towards more quanty type stuff which I like

There's nothing like real money on the line to crystalize and clarify. Most people want to solve a problem and move on; I want to make sure there isn't going to be another one! I loathe the strategic PP because the risk is too high; only recently do I "see" the exact specifics.

Anyway, I think that omitting cash for these "risk parity" allocations omits one of the economic environments. So what is a way to deal with it in lieu of duration? All I can think of is T-Bill futures.

Posted: **Fri Mar 15, 2013 12:21 am**

MachineGhost wrote:
melveyr wrote: I have noticed your posts moving towards more quanty type stuff which I like
There's nothing like real money on the line to crystalize and clarify. Most people want to solve a problem and move on; I want to make sure there isn't going to be another one! I loathe the strategic PP because the risk is too high; only recently do I "see" the exact specifics.

Anyway, I think that omitting cash for these "risk parity" allocations omits one of the economic environments. So what is a way to deal with it in lieu of duration? All I can think of is T-Bill futures.

I think the real issue is that leverage is essentially like being "short cash" because the price of leverage is based off of the short term interest rate. When I am trying to backtest a leveraged strategy I just plug a negative number into cash to make the allocation equal 100%. So there is no way to lever up cash.

If you want to magnify the returns of your portfolio through its use than you just have to pray that any "tight money" recession will be short lived. Even the PP is vulnerable to tight money recessions. Luckily we have easy money with no end sight

Posted: **Fri Mar 15, 2013 11:51 pm**

MachineGhost wrote: Ag, I thiink you're confusing the two definitions of beta. It means both the traditional correlated volatility to an index that is standard in CAPM as well as the more recent definition for exposure to any asset's risk. I do not know why they conflated the two definitions in one term. Typical "ivory tower" academics, probably.

Perhaps I am. But all the papers on MPT I have read covering several decades have all used "risk" as the theoretical term, and when applying MPT principles to the market there is no way to measure risk to come up with a number. Most of the papers then have explicitly adopted the standard measure of beta as a stand in for that unquantifiable risk in the equations.

If you read this paper on the All Weather approach, it is based on (or developed based on) MPT, and seems to use beta in the exact same way.

Posted: **Fri Mar 15, 2013 11:54 pm**

MachineGhost wrote: There's nothing engineered about the AWP other than they use leveraging and deleveraging to normalize the risk across all the asset classes relative to each other....

The problem is I'm not quite convinced us mere mortals have the ability to delevarage gold, stocks and bonds and leverage up cash so all can meet in the sweet spot.

That is exactly the engineering I am skeptical of as well. With as many assumptions as are necessary to populate the equations, it is impossible to prove they are all valid for the past, much less will hold that validity when needed in the future. (Look, is that a black swan?!) Garbage in, garbage out.

Posted: **Fri Mar 15, 2013 11:55 pm**

melveyr wrote:you can also look at a portfolios "risk allocation." To do this for each asset you look at the weighted beta of each asset in the portfolio. In this context, the individual assets beta is with respect to the total portfolio

See, there again is that beta as a substitute for risk...

Posted: **Sat Mar 16, 2013 2:00 am**

Dylan Grice, an investment strategist formerly with Societe Generale is now with Edelweiss Holdings. He writes in a recent Edelweiss Journal:

DylanGrice wrote:Of the many elemental flaws in macroeconomic practice is the true observation that the economic variables in which we might be most interested happen to be those which lend themselves least to measurement. Thus, the statistics which we take for granted and band around freely with each other measuring such ostensibly simple concepts as inflation, wealth, capital and debt, in fact involve all sorts of hidden assumptions, short-cuts and qualifications. So many, indeed, as to render reliance on them without respect for their limitations a very dangerous thing to do. As an example, consider the damage caused by banks to themselves and others by mistaking price volatility (measurable) with risk (unmeasurable). Yet faith in false precision seems to us to be one of the many imperfections our species is cursed with.

The previous issue was entitled On measuring the unmeasurable and alluded to the same concept. It was excerpted from their annual letter to shareholders, and unattributed in the Journal. I do not believe Dylan wrote it, but perhaps he did.

Posted: **Sat Mar 16, 2013 3:42 pm**

AgAuMoney wrote:
melveyr wrote:you can also look at a portfolios "risk allocation." To do this for each asset you look at the weighted beta of each asset in the portfolio. In this context, the individual assets beta is with respect to the total portfolio
See, there again is that beta as a substitute for risk...

I agree that beta is a mere proxy for risk. However, I would rather use a proxy than nothing at all! I like using quantitative measures to help verify a theoretical framework (like the PP) or use the quantitative measures to help arrive at new ideas. Whenever I go on a number crunching binge I always learn more about markets.

Posted: **Sat Mar 16, 2013 4:22 pm**

Using a proxy allows computing results, with implied correctness to those results, when that correctness is entirely dependent on the correctness of the proxy.

I would argue that using an incorrect proxy is worse than nothing at all.

Posted: **Sun Mar 17, 2013 3:47 am**

I am happy to see renewed interest in AWP. I tried to bring it up in the VP forum last year - but it didn't go anywhere...
AWP is, of course, based on PP, but with a few key improvements. And I think you can only gain, a lot, if you understand these mechanisms and have the skill set to apply them to your portfolio management:

1. The Holy Grail. The idea of having more, uncorrelated ACs is that, the more ACs you add to your portfolio, the less volatile your portfolio. This is what Ray Dalio calls the “Holy Grail”? of investing (see his interview in Schwagger’s “Hedge Fund Wizards”? for a complete explanation). In RD’s case, he has the ability to use much more uncorrelated ACs than us, the public. But still, because of the ETFs bonanza of these times we can use ACs like emerging market debt, high yield, TIPs, commodities, all unavailable to us in HBs times. And I tested that they really reduce the portfolio volatility while improving its returns.

2. Explicit Risk Parity. Allocate ACs so that each contribute equally to portfolio risk. Here is a simple mechanism for this: If all assets are somewhat uncorrelated (based on their different response to growth/inflation shocks), allocating each asset inverse proportional with their volatility would achieve this. This is called “naïve”? risk parity, as it is obviously an approximate derivation from the portfolio variance formula, but this is what is mostly used in practice. The idea is, to reallocate (or re-balance in PP parlance) on a periodic basis, but based on volatility and not just equal weights. Try to use volatile ACs, so you do not need leverage.
Now, HBPP achieves this implicitly. The 3 ACs in HBPP all have similar volatility. For instance, this is the reason he uses the long bond instead of the TNote, to match stocks and gold volatility. AWP just makes this allocation mechanism explicit.

3. Targeted Volatility. Basically, you measure the current portfolio volatility, periodically (say, once a month) and adjust the AC weights to reduce the whole portfolio volatility if above target. Or leverage if below target (but you won’t do that in practice). Adjustable target volatility is the system parameter you can dial up/down to the “sleep well at night”? value.

4. Another key thing RD does is that he overlays a risk management system on top of all this. For instance, in 2008 he reduced massively the AWP allocations based on what he called his “depression gauge”?. This is on top of the targeted volatility risk management .
RD I think uses fundamental research for his “gauge”? but you can achieve similar results using technical indicators, like MG apparently tried to show here.

Posted: **Sun Mar 17, 2013 6:48 am**

Stefan wrote: But still, because of the ETFs bonanza of these times we can use ACs like emerging market debt, high yield, TIPs, commodities, all unavailable to us in HBs times. And I tested that they really reduce the portfolio volatility while improving its returns.

So this is essentially slicing, dicing and tilting the four assets?

This is called “naïve”? risk parity, as it is obviously an approximate derivation from the portfolio variance formula, but this is what is mostly used in practice. The idea is, to reallocate (or re-balance in PP parlance) on a periodic basis, but based on volatility and not just equal weights. Try to use volatile ACs, so you do not need leverage.

Again, what do you do about cash? You can't naively risk cash because it will overwhelm the other three assets. On the other hand, you can't naively risk the other three assets and place the gap into cash either because their percentages are way too small. If you decide to scale naive risk non-cash upwards proportionally, then you get into issues of how far to go.

Now, HBPP achieves this implicitly. The 3 ACs in HBPP all have similar volatility. For instance, this is the reason he uses the long bond instead of the TNote, to match stocks and gold volatility. AWP just makes this allocation mechanism explicit.

I would argue the HBPP does it very poorly and inefficiently. As melveyr quanted earlier, gold's volatility dominates.

My solution posted elsewhere was to give up on tactical allocation and increase cash and reduce the other three assets proportionaly for my targeted risk. But I still feel like this is putting lipstick on a pig... perhaps the solution is to preserve the proportions between three risk normalized assets and then increase or decrease cash to get to the targeted volatility?

Posted: **Sun Mar 17, 2013 12:41 pm**

Stefan wrote: I am happy to see renewed interest in AWP. I tried to bring it up in the VP forum last year - but it didn't go anywhere...
AWP is, of course, based on PP, but with a few key improvements. And I think you can only gain, a lot, if you understand these mechanisms and have the skill set to apply them to your portfolio management:

1. The Holy Grail. The idea of having more, uncorrelated ACs is that, the more ACs you add to your portfolio, the less volatile your portfolio. This is what Ray Dalio calls the “Holy Grail”? of investing (see his interview in Schwagger’s “Hedge Fund Wizards”? for a complete explanation). In RD’s case, he has the ability to use much more uncorrelated ACs than us, the public. But still, because of the ETFs bonanza of these times we can use ACs like emerging market debt, high yield, TIPs, commodities, all unavailable to us in HBs times. And I tested that they really reduce the portfolio volatility while improving its returns.

2. Explicit Risk Parity. Allocate ACs so that each contribute equally to portfolio risk. Here is a simple mechanism for this: If all assets are somewhat uncorrelated (based on their different response to growth/inflation shocks), allocating each asset inverse proportional with their volatility would achieve this. This is called “naïve”? risk parity, as it is obviously an approximate derivation from the portfolio variance formula, but this is what is mostly used in practice. The idea is, to reallocate (or re-balance in PP parlance) on a periodic basis, but based on volatility and not just equal weights. Try to use volatile ACs, so you do not need leverage.
Now, HBPP achieves this implicitly. The 3 ACs in HBPP all have similar volatility. For instance, this is the reason he uses the long bond instead of the TNote, to match stocks and gold volatility. AWP just makes this allocation mechanism explicit.

3. Targeted Volatility. Basically, you measure the current portfolio volatility, periodically (say, once a month) and adjust the AC weights to reduce the whole portfolio volatility if above target. Or leverage if below target (but you won’t do that in practice). Adjustable target volatility is the system parameter you can dial up/down to the “sleep well at night”? value.

4. Another key thing RD does is that he overlays a risk management system on top of all this. For instance, in 2008 he reduced massively the AWP allocations based on what he called his “depression gauge”?. This is on top of the targeted volatility risk management .
RD I think uses fundamental research for his “gauge”? but you can achieve similar results using technical indicators, like MG apparently tried to show here.

Hey Stefan,

Thanks for the thoughtful post. I totally agree with your breakdown of the AWP and PP! It is exciting that Vanguard is coming out with a hedged international bond fund soon... I think that will make it easier to replicate the AWP portfolio for us retail investors. I want to see how it trades of course but I am thinking that

30% US Total Bond Market
30% Hedged International Bond Market
10% US Stocks
10% International Stocks
20% Gold

will be an interesting portfolio to serve as a global PP for US investors.

Or ofcourse you could do the 1/vol weightings for the bonds, stocks, and gold buckets respectively. Regardless, we are getting close to being able to replicate Dalio (except without the explicit leverage)!

Posted: **Sun Mar 17, 2013 12:42 pm**

MG,

Cash is not an AC in AWP. Cash is where you keep your capital after the AWP volatility requirements have been achieved.
So, cash volatility doesn't count in AWP and doesn't overwhelm the portfolio. You need to read the AWP paper from Bridgewater to see which are the classes they use.

This is not slicing and dicing the 4 ACs. The new ACs in Dalio's portfolio have different properties.
For instance: if the economy grows with inflation, both stocks and T-bonds suffer, but EMBs and HY and commodities do well. So, you have other classes outside gold, uncorrelated (which is key) with gold, to grow your portfolio faster with reduced volatility.

Lastly, I used the attribute "naive" instead of "simplified", but the context may have been lost here. It is "naive" in the sense that it aproximates the portfolio variance calculation. Think of the portfolio variance formula (w/o cash), think of the lack of fundamental factors response correlations among the AC set selected for AWP and look at the "simplified" formula I mentioned. It will be obvious what I meant by "naive" - in this context - .

Posted: **Sun Mar 17, 2013 2:01 pm**

Hello melveyr,

I see that you have no cash in your allocation. In AWP cash is used to modulate the total portfolio volatility so that it remains below a pre-defined target. Modulation is key here: look at my post above about the mechanism for targeted volatility to see how this is achieved.
RD actually customizes the AWP for each client based on the client's preffered TV.
By using fixed allocations, as you do, you miss this essential risk management mechanism and your portfolio volatility is, for lack of a better word, out of control

.

Also, I would look at EMBs, HYs, Commodities and TIPs as additional ACs. They have a different response than HBPP 4 ACs to growth/inflation shocks and help decrease portfolio volatility as well.

I don't think the volatility management AWP achieves is well understood here

... But once you see it, in your own backtests, you will get to appreciate it, a lot!

Posted: **Sun Mar 17, 2013 10:30 pm**

melveyr wrote: For the 1978-2011 period the following dollar weightings were what provided equal risk contribution between the assets.

Stocks: 34%
LTT: 37%
Gold: 28%

Could you post the formula you're using to normalize the risk among the three assets? I've tried many approaches but none quite seem to do the job. It should be simple but it escapes me.

Posted: **Mon Mar 18, 2013 1:09 am**

MachineGhost wrote:
melveyr wrote: For the 1978-2011 period the following dollar weightings were what provided equal risk contribution between the assets.

Stocks: 34%
LTT: 37%
Gold: 28%
Could you post the formula you're using to normalize the risk among the three assets? I've tried many approaches but none quite seem to do the job. It should be simple but it escapes me.

You managed to get the formula that outputs the risk allocations though correct? I posted how to do that part in my earlier post.

If so, make a cell that is the stdev of the risk weightings. I then used the excel Solver to minimize that cell (because the standard deviation of the risk weightings would be zero if they were all the same) with the constraint that the dollar allocations all add up to 100%.

You might get a slightly different result from me if you are using simba's spreadsheet because I was looking from 1978-2012 inflation adjusted returns and also simba's LTT are not long enough duration.

Posted: **Mon Mar 18, 2013 1:41 am**

Risk Parity is an active investing strategy - like PP - but instead of rebalancing to fixed weights based on re-balancing bands you rebalance periodically inversely proportional to the N days trailing volatility of returns of each AC.
And do not use Cash as an AC.

Posted: **Mon Mar 18, 2013 1:52 am**

Stefan wrote: You may want to change weights periodically based on N days trailing volatility of each AC.

Yeah I have considered doing that even with the PP. Luckily right now GLD, TLT, and VTI have all had roughly the same volatility so I think the PP has been close to perfect if you are ignoring correlation coefficients and only looking at vol. However, during the gold bubble of the late 70s gold was way more volatile than the other components before it popped. Looking at trailing vol could have helped mitigated the huge PP drawdowns during that period.

PS:

Stefan, I know more about Risk Parity than you are giving me credit for

I just value simplicity as well.

Posted: **Mon Mar 18, 2013 2:12 am**

melveyr - I appreciate your input and your posts have been one of the reasons I decided to join the conversation.

Unfortunately, there is no simple solution here.

AWP is superior to PP for the reasons I wrote. It is like comparing windows 95 to windows 8 (or, maybe , should I stick with win 7

), different generations of the same great concept, with AWP adding state of the art research of volatility based risk management to the basic core system. Something PP completely lacks and gets only implicitly,and "by luck", as you mentioned.

So, I think, it is worth exploring what Ray Dalio's AWP does, trying to repro it and comparing it with the PP performance over the same period and see if AWP provides a better performance with much less risk.

And there is no simple or complex way to do that. We just need to reverse engineer the known/published AWP mechanisms and see where it gets us.

Posted: **Mon Mar 18, 2013 2:23 am**

I'm not using Excel, but coding this up to be dynamic in real time as part of a trading system, hence I don't have Solver available to brute force an optimal solution with contraints. For now, I think it works reasonbly well to use 25% as the initial value weight and after normalizing the risk of the three assets, using the adjustment factors to modify the 25% value weight and then using a separate beta factor to manually increase or decrease the resulting value weights to get to the portfolio risk target. A problem is if I use bonds as the risk to normalize to, then it will always be a 25% value weight in the portfolio unless I use a higher or lower risk that doesn't match the assets. That doesn't seem elegant. What also bothers me is I rather have the whole process be adaptive and decide the optimal weights by itself without having to first use a fixed 25% value weight or adjust the manual beta factor.

But as far as this implementation goes for bond's risk, the portfolio return is 7.88% CAR and -7.24% MaxDD. Bond's B&H is 8.20% and -28.28% MaxDD. I can use a beta factor of 1.44x to get the portfolio to my risk target of 8.31% CAR and -9.95% MaxDD. This is not bad at all for not using any market timing or tactical allocation. The beta factored weights it picks as of 2/15/2013 was:

Stocks: 19.29%
Bonds: 36%
Gold: 6.77%
Cash: 37.91%

I see implementation difficulties with slicing and dicing to AWP, but I relish the challenge. Something like that "portfolio at once" broker would be necessary to keep the rebalancing costs down. I doubt that Fidelity or Schwab have enough free niche ETF's.

There seems to be limited information availalbe as to why AWP selects things like emerging market bonds and high yield bonds, or the criteria needed to choose them. If they do good in inflation, it would preclude anything but ultra-short maturities, for instance.

Posted: **Mon Mar 18, 2013 2:52 am**

From the Bridgewater All Weather Story paper, the ACs list for the 4 different regimes:

I would add gold & real estate to this set and use HY for Corporate Credit. And that should be it.

Gyroscopic Investing

Engineering Targeted Returns and Risk

Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk

Re: Engineering Targeted Returns and Risk