Engineering Targeted Returns and Risk

[Stefan]

Interesting.  What is responsible for the lesser max drawdown of the AWP vs PP in 2008?  Your tactical allocation or the asset composition?

Actually both.

In their letters, Bridgewater describes their safeguard mechanism for protecting their investors in a depressionary environment. This is an environment in which, to quote the master:

... entities are forced to sell assets in order to pay down debt, reducing the value of those assets and increasing the need to delever even further. As a result, severe credit and liquidity problems arise and the financial and economic system ceases to function normally. These conditions are often self reinforcing, creating the possibility of severe and prolonged underperformance of asset classes with equity or credit related risks.

Or, what we would call today a Risk Off regime.

So, in order to preserve capital in that environment they designed a "Safe Portfolio" made of: T-Bonds, IL Bonds, Gold and TBills. The SP is designed to be imune to recession,depression, inflation and currency debasement.

They detect the Risk Off environment with a "Depression Gauge". And they transition AWP to the "Safe Portfolio" incrementally, based on this indicator. And gradually back to AWP, as the environment shifts back to Risk On.

Here is when they applied the shifts back and forth in 2008-2009. (I think the 10% AWP in the chart title stands for 10% volatility target).
So, they gradually moved to SP from AWP in early 2008 and transitioned gradually out in early 2009.


To provide a sense for how bad a depression environment can be for a portfolio of asset classes, in the following chart they show the returns of a conventional (65/35) asset mix, All Weather, and the Safe Portfolio during the Great Depression. While the All Weather mix did much better than the equity-heavy conventional portfolio, it still experienced materially negative returns. However, the Safe Portfolio remained relatively well insulated during this period


In the backtest I showed, I just created my own Risk On/Risk Off indicator - based on technical measures - and I tried to reproduce their results. However, I was much more aggresive than Bridgewater in shifting between Risk On/Off (though still using incremental moves). This explains the low drawdown in 2008 in my AWP implementation backtest vs the PRPFX as well as vs the Bridgewater AWP published equity curve in the pic above.

[MachineGhost]

In the backtest I showed, I just created my own Risk On/Risk Off indicator - based on technical measures - and I tried to reproduce their results. However, I was much more aggresive than Bridgewater in shifting between Risk On/Off (though still using incremental moves). This explains the low drawdown in 2008 in my AWP implementation backtest vs the PRPFX as well as vs the Bridgewater AWP published equity curve in the pic above.

Are you deducting commissions and taxes to account for the increased turnover?

[MachineGhost]

Here is my "risk parity" version of the PP using the standard assets, except using 3-year CM Treasuries to imitate a 5-year ladder:

[align=center]1975-2013.29: 8.87% CAGR, -17.16% MaxDD
[/align]

And my "targeted risk" version of the above that I am currently in the process of implementing:

[align=center]1975-2013.29: 8.18% CAGR, -9.98% MaxDD
[/align]

I will have two PP's.  The first, bigger one will implement the above concepts with direct ownership.  And the second, smaller one (not a VP) will implement the rest of the AWP concepts with Schwab's free ETF's if I can get the data and see how well it backtests, etc..  Finding long-run history on a daily granularity for "slicing and dicing" is bound to be a problem.

[gap]

Stephan:

I have to be very skeptical of a "depression gauge". It is easy to define one looking back and it is easy to put together a portfolio that does well for that period. The challenge for any one is to  define clearly and measurably one that defines that period using only data available prior to the so called  "depression". The real test will be to see if it works for the next "depression".

I am also wary of the many press releases and interviews by RD over the last year or so since RD is trying to sell his company.

RD is very good in his profession, but a vague "depression gauge" smacks a little bit of sophistry IMO.

[Stefan]

Are you deducting commissions and taxes to account for the increased turnover?

No, I don't. Taxes may affect AWP CAGR, indeed, as the portfolio is heavy with income generators. Comissions are a problem only if you have a small account.

About your backtests:

First off, I am glad to meet another AB user. Isn't it a great platform ?

It would be also interesting to see a plot of the equity curve.

About the implementation: do you manage risk using "targeted volatility" or some other concepts? Also, are you thinking of including a Risk On/Risk Off as an additional risk management mechanism?

[Stefan]

Stephan:

I have to be very skeptical of a "depression gauge". It is easy to define one looking back and it is easy to put together a portfolio that does well for that period. The challenge for any one is to  define clearly and measurably one that defines that period using only data available prior to the so called  "depression". The real test will be to see if it works for the next "depression".

I am also wary of the many press releases and interviews by RD over the last year or so since RD is trying to sell his company.

RD is very good in his profession, but a vague "depression gauge" smacks a little bit of sophistry IMO.

I read most of their research & white papers and I think they are one of the best at understanding how the economic machine works. With this understanding of the machine and with their huge resources, I think it is probable that they developed a pretty good model of the machine to work with.

About Dalio's interviews, I always found them very enlightening. His is extremely clear,  to the point and, mostly, conveys ideas and concepts you didnt know and you could use. It is super useful to read Schwagger's interview with Dalio in "Hedge Fund Wizards" to see the quality of the content provided by RD and compare that with other money managers interviews.

[MachineGhost]

First off, I am glad to meet another AB user. Isn't it a great platform ?

I have a love-hate relationship with it.  Mostly because minor things are overlooked in being implemented because it is a one-man programming operation and he prioritizes weirdly.  Still, I can't imagine trusting anything else...  the rest are far worse.

About the implementation: do you manage risk using "targeted volatility" or some other concepts? Also, are you thinking of including a Risk On/Risk Off as an additional risk management mechanism?

I'm a drawdown fan so I'm not using volatility at all.  I am burned out on tactical allocation, so thats very low on my priority list, but a leopard can't change his spots.  For now, I need to focus on implementing the AWP.  I think there are practical difficulties in converting a portfolio that is held directly for maximum safety to a "risk off" portfolio anyway.  Not something I want to relish dealing with right now.

[MachineGhost]

I was curious how much risk on or risk off the equal weight PP was compared to the risk parity PP.  Turns out the equal weight is leveraged by around 3.5%, i.e. I have to leverage the cashless risk parity by 3.5% to an equivalent equal weight PP portfolio.

[MachineGhost]

The risk of the PP is sensitive to the rebalancing date.  To wit:

5/1/73: -25.25% MaxDD
6/1/73: -24.27% MaxDD
7/1/73: -24.17% MaxDD
8/1/73: -24.17% MaxDD
9/1/73: -22.96% MaxDD
10/1/73: -22.33% MaxDD
11/1/73: -21.29% MaxDD
12/1/73: -20.78% MaxDD
1/1/74: -17.94% MaxDD
2/1/74: -20.07% MaxDD
3/1/74: -22.20% MaxDD
4/1/74: -25.66% MaxDD

Since the PP is overweight to gold's risk, this is directly attributable to gold's bubble peak and subsequent collapse in early 1980.

[Stefan]

The risk of the PP is sensitive to the rebalancing date.  To wit:

5/1/73: -25.25% MaxDD
6/1/73: -24.27% MaxDD
7/1/73: -24.17% MaxDD
8/1/73: -24.17% MaxDD
9/1/73: -22.96% MaxDD
10/1/73: -22.33% MaxDD
11/1/73: -21.29% MaxDD
12/1/73: -20.78% MaxDD
1/1/74: -17.94% MaxDD
2/1/74: -20.07% MaxDD
3/1/74: -22.20% MaxDD
4/1/74: -25.66% MaxDD

Since the PP is overweight to gold's risk, this is directly attributable to gold's bubble peak and subsequent collapse in early 1980.

A few notes on what a "pure" PP investor should expect (to endure), related to MaxDD and under-performance:

The PP proxy, PRPFX, dropped -27% in 2008. This time it was the stocks collapse. And MaxDDs of this magnitude will continue to happen every time in the future when (not if) one of the 3 asset classes go through a bubble followed by a collapse.

For (most) investors, volatility of returns is the key criteria to use or abandon a strategy, fund or money manager. They will put up with underperforming (for a short time, see below) if the maxDD is quite mild and if, when the market tanks, the strategy protects them, maintaining the same low volatility. But the pain of a -25% MaxDD is not acceptable to most investors, especially those close to retirement.

Now, speaking of underperformance, I think most PPers will have a very hard time to stick to the strategy in a powerful bull market (bubble) for stocks like 1995-2000:



For the period, PRPFX averaged a 5% CAGR vs 26% for SPY.

[Pointedstick]

I don't think that PRPFX is a good proxy for the PP anymore, and it really showed in 2008. It's too overweight on inflation-linked assets and didn't have enough long treasuries to prevent that brutal loss. And its Swiss Francs are wasted space.

[Stefan]

I don't think that PRPFX is a good proxy for the PP anymore, and it really showed in 2008. It's too overweight on inflation-linked assets and didn't have enough long treasuries to prevent that brutal loss. And its Swiss Francs are wasted space.

Do you have the PP MaxDD in 2008 and the CAGR for 1995-2000?

Using simba spreadsheet you get 5.5% CAGR for PP vs 5% for PRPFX over 1995-2000.

[melveyr]

PRPFX is not a good proxy for the 25 x 4 portfolio. Max drawdown for the PP in 2008 was around 11.9%. If you go back a little bit farther looking at 2007-2008 than the max drawdown was 14.9%.

[Stefan]

PRPFX is not a good proxy for the 25 x 4 portfolio. Max drawdown for the PP in 2008 was around 11.9%. If you go back a little bit farther looking at 2007-2008 than the max drawdown was 14.9%.

Was that end of day or end of month? PRPFX MaxDD EOM was -19% vs -27% EOD

[melveyr]

PRPFX is not a good proxy for the 25 x 4 portfolio. Max drawdown for the PP in 2008 was around 11.9%. If you go back a little bit farther looking at 2007-2008 than the max drawdown was 14.9%.

Was that end of day or end of month? PRPFX MaxDD EOM was -19% vs -27% EOD

I used ETF replay which is EOD.

[MachineGhost]

Hmm, I get 8.23% CAGR and -5.49% MaxDD between 4/1994 to 4/2000 with the vanilla PP.  The stocks allocation only got 5.85% CAGR.

[MachineGhost]

That's weird.  Shiller data gives something close to Stefan's 20% for stocks.  Does yours include dividends?

Total return all the way, baby!

[MachineGhost]

Ught!

What Happens when Risk Parity Divorces the Long Bond?
Summary

? Risk parity (“RP”?) shifts the weights between different asset classes so that each asset provides the same amount of volatility (measured by standard deviation in our case) as other assets.

? A common critique of risk parity strategies is their heavy exposure to the long-bond—an asset class with an excellent  run  over  the  past  30  years.  To  many,  any  strategy  that  allocated  heavily  to  long-bonds—be  it skilled-based allocations or mere luck—was a sure-fire winner.

Empirical Question: What happens when the long-bond is no longer an asset class?

? Hypothesis 1: If risk parity adds asset allocation value, we should see outperformance with or without the inclusion of long-bonds as an asset class.

Empirical Results:

? Risk parity performance is driven by heavy allocations to long-bonds. When bonds are not an asset class, there is little evidence from our tests that it an effective asset allocation tool.

? The simple equal-weight asset allocation scheme with a moving average trading rule outperforms risk parity.

[MachineGhost]

If you're coming back to tactical allocation, you might want to look at the Butler and Philbrick stuff again.

http://www.advisorperspectives.com/dsho ... panese.php

It's results pretty much bear out the same with the Japan data.

Interesting article, but it's not an answer to risk parity being over-exposed to fixed income.  However, this article by the same authors discusses the problem: http://www.advisorperspectives.com/dsho ... -Prime.php

It seems to me there is no better risk-reward portfolio that isn't going to involve more active management than any of us would like.

I think its important to point out that what Wall Street is currently in the fad of implementing is static risk parity and levering it up to reach a target risk.  Static in the manner of the weights not changing according to short-term risk considerations any more than a strategic allocation does, so both suffer from the same flaw.  And, being overweight to fixed income, it is likely to be disastrous when long-term yields or inflation rises as sophie and others have noted.  My implementation of risk parity is adaptive at each annual rebalancing, except that I'm not using short-term risk measurements except in the beginning years, so it too seems to suffer from the same flaw.  Yet, I am skeptical increased rebalancing frequency along with short-term risk measurements will result in a net increase after transaction costs and taxes.

So excluding TA and momentum, what I've read so far seems to point to using an adaptive risk parity at both the asset and portfolio level, but also taking asset correlation into account over the short-term, is the best way to go.  In other words, you can just run a mean-variance optimization on the PP at every rebalancing.  This does make sense because the PP assumes long-term asset correlations that widely diverge in the short-term which eats away at your return.  The definitive question seems to be: how short-term is short-term to rebalance over?

[Stefan]

So excluding TA and momentum, what I've read so far seems to point to using an adaptive risk parity at both the asset and portfolio level, but also taking asset correlation into account over the short-term, is the best way to go.  In other words, you can just run a mean-variance optimization on the PP at every rebalancing.  This does make sense because the PP assumes long-term asset correlations that widely diverge in the short-term which eats away at your return.  The definitive question seems to be: how short-term is short-term to rebalance over?

I do not see a risk problem associated with fixed income exposure when the long bond is in a bear market, in the context of an AAA system, like the one described in the paper, which uses momentum for relative strength ranking - with monthly rebalancing. Do you?

[MachineGhost]

I do not see a risk problem associated with fixed income exposure when the long bond is in a bear market, in the context of an AAA system, like the one described in the paper, which uses momentum for relative strength ranking - with monthly rebalancing. Do you?

No.  The key seems to be using a short enough lookback period so that the rebalancing will properly adjust when volatility increases.  Would a bond bear market increase volatility or would it be death by a thousand cuts?

[Stefan]

I do not see a risk problem associated with fixed income exposure when the long bond is in a bear market, in the context of an AAA system, like the one described in the paper, which uses momentum for relative strength ranking - with monthly rebalancing. Do you?

No.  The key seems to be using a short enough lookback period so that the rebalancing will properly adjust when volatility increases.  Would a bond bear market increase volatility or would it be death by a thousand cuts?

In addition, you could use an absolute momentum filter. For instance: you allocate capital to TBonds only if 6 month return on TBonds > 6 month TBills return. And you test this condition every month, using this 6 month rolling lookback window. This condition will not be true as long as TBonds are in a bear market => so TBonds won't increase the portfolio  volatility nor will they whipsaw you.

[MachineGhost]

Been doing backtesting on different weighting schemes for the PP.  Here are the results so far:

Code: Select all

Annual Rebalancing (6/1974+)

Equal Weight 3x33		8.60% CAR	-29.84% MaxDD
Volatility Weight (1%)		9.08% CAR	-26.00% MaxDD
Equal Weight 4x25		8.08% CAR	-24.32% MaxDD
Momo, Equal Weight		9.10% CAR	-24.03% MaxDD
Momo, Volatility Weight (1%)	9.23% CAR	-22.26% MaxDD
Volatility Weight (Scaled)	8.37% CAR	-20.16% MaxDD

Quarterly Rebalancing (6/1974+)

Equal Weight 3x33		5.88% CAR	-30.71% MaxDD
Volatility Weight (1%)		5.61% CAR	-16.48% MaxDD
Equal Weight 4x25		5.42% CAR	-20.62% MaxDD
Momo, Equal Weight		6.70% CAR	-29.17% MaxDD
Momo, Volatility Weight (1%)	6.33% CAR	-22.44% MaxDD
Volatility Weight (Scaled)	5.15% CAR	-22.52% MaxDD

I used top 2 (out of 3) for momo and .39% per trade for transactions costs (commissions, spread, market impact).

[Stefan]

Been doing backtesting on different weighting schemes for the PP.  Here are the results so far:

Code: Select all

Annual Rebalancing (6/1974+)

Equal Weight			8.60% CAR	-29.84% MaxDD
Volatility Weight (1%)		9.08% CAR	-26.00% MaxDD
Equal Weight (w/Cash)		8.08% CAR	-24.32% MaxDD
Momo, Equal Weight		9.10% CAR	-24.03% MaxDD
Momo, Volatility Weight (1%)	9.23% CAR	-22.26% MaxDD
Volatility Weight (Scaled)	8.37% CAR	-20.16% MaxDD

Quarterly Rebalancing (6/1974+)

Equal Weight			5.88% CAR	-30.71% MaxDD
Volatility Weight (1%)		5.61% CAR	-16.48% MaxDD
Equal Weight (w/Cash)		5.42% CAR	-20.62% MaxDD
Momo, Equal Weight		6.70% CAR	-29.17% MaxDD
Momo, Volatility Weight (1%)	6.33% CAR	-22.44% MaxDD
Volatility Weight (Scaled)	5.15% CAR	-22.52% MaxDD

I used top 2 (out of 3) for momo and .39% per trade for transactions costs (commissions, spread, market impact).

I think the MaxDD range makes all these systems very hard to trade. In addition, the upside is very limited. If you use the equal weight annually rebalanced PP as a benchmark, the systems shown can barely match this upside, while capturing most of the downside as well. I do not see a real advantage in using these as opposed to the standard PP.

To make a qualitative difference, the system should manage risk: both moderate it and contain it in times of financial panic. Do that and the upside will take care of itself.
There is a difference between risk moderation, which you show here, using volatility weighting, and risk containment, which avoids any all big drawdowns. There are a few ways these 2 can be achieved, even with the PP portfolio, and I presented a few methods and the resultant backtest numbers in my posts.

To synthesize what I already said in this thread, there are 2 forms of risk management to be overlayed on top of PP (or AWP) to make the system more palatable to the conservative investor:

1.  Risk Moderation  achievable by :
            a.    Using a risk parity allocation formula based on volatility weights
            b.    Managing the whole portfolio volatility to a target.

2.  Risk Containment achievable by:
          a.   Asset classes trend. Capturing most of the trend upside and avoiding most of the trend downside.
          b.   Minimizing exposure gradually in times of market turbulence.

Here is an AWP example which makes use of most of these techniques (vs PRPFX as a benchmark):
The CAGR is 11.23% and MaxDD -6%

[MachineGhost]

1.  Risk Moderation  achievable by :
            a.    Using a risk parity allocation formula based on volatility weights
            b.    Managing the whole portfolio volatility to a target.

How does one align the whole portfolio to a target after scaling each asset's volatility to a 1% daily target?  It seems like this would be easy to do after the new weights and equity curve are generated, but not before a rebalancing.

2.  Risk Containment achievable by:
          a.   Asset classes trend. Capturing most of the trend upside and avoiding most of the trend downside.
          b.   Minimizing exposure gradually in times of market turbulence.

So essentially TA is required here.  Momo is not enough.

Here is an AWP example which makes use of most of these techniques (vs PRPFX as a benchmark):
The CAGR is 11.23% and MaxDD -6%

I assume your portfolio involves more than the standard four PP assets?  The momo would do better if more assets than the standard three were included.  I suspect the same applies to yours.