Investing

Portfolio Allocation

  • Last Updated:
  • Jul 18th, 2016 7:44 am
Tags:
None
Deal Addict
Apr 22, 2014
3097 posts
474 upvotes
Oshawa, ON

Portfolio Allocation

How do you guys allocate holdings in your portfolio?

1/n? market cap? efficient frontier/max sharpe ratio? risk parity/equal risk? 1/sigma? gut feel?

Equal risk concentration is new to me and I was stunned by the results vs. a market cap weight vs. max sharpe ratio is pretty impressive. Combined with a momentum indicator it takes you through 2008 with a small gain (although it under performs in 2015 by a couple %) on a US sector allocation portfolio (that'd be an alternative to someone whose portfolio is US equity/bond, but does require a move in to gold at some points).

Some recent One portfolio I manage with some 9 ETFs (Canada equity/bond/REIT, US equity/bond, EAFE equity/bond, EM equity, gold) is indicating that holding only 100% the Canada REIT is an efficient portfolio according to Markowitz. Max Sharpe ratio excludes US equity right now.

I spent a lot of time on this years ago with no money at play but recently have had to get my models caught up again because there's real money at play now, so it's top of mind.
19 replies
Deal Addict
Apr 22, 2014
3097 posts
474 upvotes
Oshawa, ON
For you couch potato-ers using the VCN/VAB/VXC mix, I ran an equal risk model and get weights 63%/71%/-34% (ie shorting the VXC).
The max sharpe ratio portfolio is 1/3 VCN and 2/3 VAB and no VXC (like bang on, to the infinite decimal place, a mathematical singularity). I get the feeling if I re-ran the equal risk model without VXC in the mix it would be similar.
A 1/sigma approach gets you 20%/60%/20%. Danger lies in disregarding correlations.
Market cap doesn't really work with the couch potatato for Canadians because you'd be at about 95% VXC. Which some theorize would be a good thing because your entire sense of existence is nearly 100% correlated with Canada.
Deal Fanatic
User avatar
Dec 14, 2010
7113 posts
9300 upvotes
eldeejay wrote: How do you guys allocate holdings in your portfolio?

1/n? market cap? efficient frontier/max sharpe ratio? risk parity/equal risk? 1/sigma? gut feel?

Equal risk concentration is new to me and I was stunned by the results vs. a market cap weight vs. max sharpe ratio is pretty impressive. Combined with a momentum indicator it takes you through 2008 with a small gain (although it under performs in 2015 by a couple %) on a US sector allocation portfolio (that'd be an alternative to someone whose portfolio is US equity/bond, but does require a move in to gold at some points).

Some recent One portfolio I manage with some 9 ETFs (Canada equity/bond/REIT, US equity/bond, EAFE equity/bond, EM equity, gold) is indicating that holding only 100% the Canada REIT is an efficient portfolio according to Markowitz. Max Sharpe ratio excludes US equity right now.

I spent a lot of time on this years ago with no money at play but recently have had to get my models caught up again because there's real money at play now, so it's top of mind.
Each individual does it differently, and depending on your goals, allocation would be different.

Speaking for myself, my goals are towards safety and consistency of dividend growth, so my portfolio is equally weighted into 10 sectors, and each sector has an equal distribution of the stocks that belongs to that sector and meet my criteria. Some sectors have lots of stocks, some just have a few. Since the sector weight is the same, there is more exposure to certain stocks than others, but the exposure from a sector perspective is equally distributed.

I have no othet type of investments (bond or gold or cash), but I do trading for the short term (which holds bonds and gold), and my trading allocation is 50% on a Canadian model, 20% each on 2 US models and 10% on a very short term US model.

Rod
Deal Addict
Dec 3, 2014
2348 posts
1840 upvotes
Ontario
eldeejay wrote: For you couch potato-ers using the VCN/VAB/VXC mix, I ran an equal risk model and get weights 63%/71%/-34% (ie shorting the VXC).
The max sharpe ratio portfolio is 1/3 VCN and 2/3 VAB and no VXC (like bang on, to the infinite decimal place, a mathematical singularity). I get the feeling if I re-ran the equal risk model without VXC in the mix it would be similar.
A 1/sigma approach gets you 20%/60%/20%. Danger lies in disregarding correlations.
Market cap doesn't really work with the couch potatato for Canadians because you'd be at about 95% VXC. Which some theorize would be a good thing because your entire sense of existence is nearly 100% correlated with Canada.
I don't couch potato per se but just buying Canada and bonds sounds like a recipe for disaster. I may be misunderstanding you though as I find your writing style to be difficult to follow.
Deal Addict
Apr 22, 2014
3097 posts
474 upvotes
Oshawa, ON
llpresident wrote: I don't couch potato per se but just buying Canada and bonds sounds like a recipe for disaster. I may be misunderstanding you though as I find your writing style to be difficult to follow.
It's just what 2 models spit out given some assumptions as of right now. You have 3 choices and 2 models say exclude 1 choice. It's based on a less than 1 month rebalancing target. You'd want to reassess this weekly to monthly.
Sr. Member
Mar 13, 2009
781 posts
212 upvotes
Maple
eldeejay wrote: For you couch potato-ers using the VCN/VAB/VXC mix, I ran an equal risk model and get weights 63%/71%/-34% (ie shorting the VXC).
The max sharpe ratio portfolio is 1/3 VCN and 2/3 VAB and no VXC (like bang on, to the infinite decimal place, a mathematical singularity). I get the feeling if I re-ran the equal risk model without VXC in the mix it would be similar.
A 1/sigma approach gets you 20%/60%/20%. Danger lies in disregarding correlations.
Market cap doesn't really work with the couch potatato for Canadians because you'd be at about 95% VXC. Which some theorize would be a good thing because your entire sense of existence is nearly 100% correlated with Canada.
Isn't not trying to time the market the whole point of couch potato?
Deal Addict
Apr 22, 2014
3097 posts
474 upvotes
Oshawa, ON
ak1004 wrote: Isn't not trying to time the market the whole point of couch potato?
Yes but...
Member
Sep 1, 2013
403 posts
97 upvotes
eldeejay wrote: How do you guys allocate holdings in your portfolio?

1/n? market cap? efficient frontier/max sharpe ratio? risk parity/equal risk? 1/sigma? gut feel?

Equal risk concentration is new to me and I was stunned by the results vs. a market cap weight vs. max sharpe ratio is pretty impressive. Combined with a momentum indicator it takes you through 2008 with a small gain (although it under performs in 2015 by a couple %) on a US sector allocation portfolio (that'd be an alternative to someone whose portfolio is US equity/bond, but does require a move in to gold at some points).

Some recent One portfolio I manage with some 9 ETFs (Canada equity/bond/REIT, US equity/bond, EAFE equity/bond, EM equity, gold) is indicating that holding only 100% the Canada REIT is an efficient portfolio according to Markowitz. Max Sharpe ratio excludes US equity right now.

I spent a lot of time on this years ago with no money at play but recently have had to get my models caught up again because there's real money at play now, so it's top of mind.
Risk parity approaches work primarily relative to a benchmark portfolio allocation. You will still need a benchmark portfolio which has a defined level of risk. For individual investors, that risk level is a function of a number of factors, such as age, liquidity needs, income levels, etc. It changes with time. Institutional investors, such as pension funds, have a rather more stable risk requirement over time and can use risk parity more effectively using long-term backtests.

The approach to the backtest you ran requires care. I know the world of academia likes to research in this fashion and publishes papers 'with 20/20 hindsight', a professional investor would have to more systematic about it. What you need to do is, transport yourself back in time - i.e. immediately before the 2008 markets - and ask yourself the question: would I have allocated to this portfolio (using this momentum indicator) given what I knew of the world at that time? Without the 2008 data point, what evidence would you have to make sure you aren't looking at it solely due to hindsight? If the 2008 market drop = a once-in-20-years event, your hindsight portfolio may well underperform in the following 19 years.
Member
Sep 1, 2013
403 posts
97 upvotes
eldeejay wrote: For you couch potato-ers using the VCN/VAB/VXC mix, I ran an equal risk model and get weights 63%/71%/-34% (ie shorting the VXC).
The max sharpe ratio portfolio is 1/3 VCN and 2/3 VAB and no VXC (like bang on, to the infinite decimal place, a mathematical singularity). I get the feeling if I re-ran the equal risk model without VXC in the mix it would be similar.
A 1/sigma approach gets you 20%/60%/20%. Danger lies in disregarding correlations.
Market cap doesn't really work with the couch potatato for Canadians because you'd be at about 95% VXC. Which some theorize would be a good thing because your entire sense of existence is nearly 100% correlated with Canada.
How were you able to get a negative allocation using a risk parity model? What was your definition of risk?
ak1004 wrote: Isn't not trying to time the market the whole point of couch potato?
Correct, but you can still have a negative allocation to something as a passive investor. You would have to be careful with the margin calls, however, or invest in an inverse ETF/asset, which - on the flip side - will carry a higher MER.
Deal Addict
Apr 22, 2014
3097 posts
474 upvotes
Oshawa, ON
Ironcat wrote: Risk parity approaches work primarily relative to a benchmark portfolio allocation. You will still need a benchmark portfolio which has a defined level of risk. For individual investors, that risk level is a function of a number of factors, such as age, liquidity needs, income levels, etc. It changes with time. Institutional investors, such as pension funds, have a rather more stable risk requirement over time and can use risk parity more effectively using long-term backtests.

The approach to the backtest you ran requires care. I know the world of academia likes to research in this fashion and publishes papers 'with 20/20 hindsight', a professional investor would have to more systematic about it. What you need to do is, transport yourself back in time - i.e. immediately before the 2008 markets - and ask yourself the question: would I have allocated to this portfolio (using this momentum indicator) given what I knew of the world at that time? Without the 2008 data point, what evidence would you have to make sure you aren't looking at it solely due to hindsight? If the 2008 market drop = a once-in-20-years event, your hindsight portfolio may well underperform in the following 19 years.
I was using equal risk allocation across MSCI US sectors plus the TLT and GLD. That doesn't need a benchmark but the benchmark to compare would be the MSCI US index plus long treasuries and gold. You just need to solve the weight so that each sector contributes equal risk. Naturally there are some assumptions going in to your variances and covariances.
Deal Addict
Apr 22, 2014
3097 posts
474 upvotes
Oshawa, ON
Ironcat wrote: How were you able to get a negative allocation using a risk parity model? What was your definition of risk?

If weights are not constrained to be greater than one but still must some to one the target function is to have weight x beta equal for all portfolio elements and sum of these must equal the portfolio vol. beta here is the beta of each relative to the portfolio not the usual beta.

For variance and covariances I calculate the matrix using from 120 days past to 30 days past then use that to do a 30 day GARCH. I wouldn't trust any of this beyond a week.


The world ex Canada has been pretty volatile recently with poor returns relative to Canada. More simply Canada's ex piste Sharpe ratio is positive and the ROW is negative for the recent history
Deal Fanatic
User avatar
Nov 2, 2013
5697 posts
1522 upvotes
Edmonton, AB
Everyone has their preference. For me much of my holdings are roughly the same market cap, so I just do equal dollar amounts to begin. I thought about doing weights based on the volatility of each stock, or standard deviation of prices (I believe there is another term for this).
Accountant (Public Practice)
Member
Sep 1, 2013
403 posts
97 upvotes
eldeejay wrote: If weights are not constrained to be greater than one but still must some to one the target function is to have weight x beta equal for all portfolio elements and sum of these must equal the portfolio vol. beta here is the beta of each relative to the portfolio not the usual beta.
I didn't really get that sentence to be honest (you might wanna use commas or actual numbers, etc.), but tell me if I'm wrong about my understanding of what you are doing.

1. You define a new metric - say, 'relative beta' - calculated as either (A) beta of each security minus portfolio beta or (B) beta of each security divided by portfolio beta. (B) would still not give you a negative weight unless beta for that security is negative in the first place. So, let's say you are using (A) to define 'relative beta' for each security.

2. You then add up (or net) all the 'relative betas' to come up with the 'denominator'. The weight for each security = its relative beta / denominator.

3. You finally extract optimal weights using a numerical method or Excel's Solver. Not quite sure though what outcome you used to define the optimal weights. Return? Sharpe ratio? Volatility? Target beta?

In any case, using just betas, you have achieved market risk parity (and not total risk parity). Using relative betas, you have achieved parity to a target portfolio beta - rather than the market risk.
Deal Addict
Apr 22, 2014
3097 posts
474 upvotes
Oshawa, ON
Ironcat wrote: I didn't really get that sentence to be honest (you might wanna use commas or actual numbers, etc.), but tell me if I'm wrong about my understanding of what you are doing.

1. You define a new metric - say, 'relative beta' - calculated as either (A) beta of each security minus portfolio beta or (B) beta of each security divided by portfolio beta. (B) would still not give you a negative weight unless beta for that security is negative in the first place. So, let's say you are using (A) to define 'relative beta' for each security.

2. You then add up (or net) all the 'relative betas' to come up with the 'denominator'. The weight for each security = its relative beta / denominator.

3. You finally extract optimal weights using a numerical method or Excel's Solver. Not quite sure though what outcome you used to define the optimal weights. Return? Sharpe ratio? Volatility? Target beta?

In any case, using just betas, you have achieved market risk parity (and not total risk parity). Using relative betas, you have achieved parity to a target portfolio beta - rather than the market risk.
I think your definition of risk parity is different than definition of equal risk weighting.

Anyway I didn't invent the method I'll post the link to the paper I took the idea from on Monday.

The criterion for risk and weighting is based on the contribution to risk of each asset In the portfolio is equal. Risk defined by weight x beta but like I said its beta relative to the portfolio ie cov(asset, portfolio) not cov(asset,index) like usual. Solve such that w x beta is equal for every constituent such that sum w = unity.

I'm using newtons method with interface in Excel and Matabele doing the heavy lifting behind that.

Please don't think I think I'm preaching the gospel here I just tried something out and thought I'd ask what the big shots here do.

I experiment with different portfolio allocation ideas to try to get to something that is intuitive and easy long run.

Mostly I believe in the 70/30 portfolio except varying the allocation from time to time depending on market conditions.
Deal Addict
User avatar
Jan 19, 2007
1451 posts
811 upvotes
Torontario
eldeejay wrote: I think your definition of risk parity is different than definition of equal risk weighting.
Anyway I didn't invent the method I'll post the link to the paper I took the idea from on Monday.

The criterion for risk and weighting is based on the contribution to risk of each asset In the portfolio is equal. Risk defined by weight x beta but like I said its beta relative to the portfolio ie cov(asset, portfolio) not cov(asset,index) like usual. Solve such that w x beta is equal for every constituent such that sum w = unity.

I'm using newtons method with interface in Excel and Matabele doing the heavy lifting behind that.

Please don't think I think I'm preaching the gospel here I just tried something out and thought I'd ask what the big shots here do.

I experiment with different portfolio allocation ideas to try to get to something that is intuitive and easy long run.

Mostly I believe in the 70/30 portfolio except varying the allocation from time to time depending on market conditions.
I might think you are being caught in the complex mechanics that finally for 80% of the retail investors do not make a material difference.
Majority of people do not answer basic questions for themselves about objectives, risk-tolerances, and do not acknowledge basic assumptions and constraints current and into the future. Not to mention ability to stick with a plan!

Examples: if you work in the financial industry, account for that the portfolio allocation as your income is an exposure. If you have a large mortgage, aren't you exposed to RE enough and how can you hedge? If you are going to retire somewhere south, are your CAD-investments over-weighted? If you are in low interest environment, are you optimizing vainly for growth or looking to cut costs (trading and management fees)?

Key thing to reflect on -- my personal thought is a lot of portfolio theories these days begin to crack in a world-wide phenomenon of approximately 10 years low-interest rate environment. Example negative yields on short and almost worthless return on long-term bonds.

Otherwise you do all funky stuff with Matlab plug-ins or @Risk Monte-Carlo simulations - such effort is only worth if one is clear on the more strategic baseline.
Member
Sep 1, 2013
403 posts
97 upvotes
eldeejay wrote: I think your definition of risk parity is different than definition of equal risk weighting.

Anyway I didn't invent the method I'll post the link to the paper I took the idea from on Monday.

The criterion for risk and weighting is based on the contribution to risk of each asset In the portfolio is equal. Risk defined by weight x beta but like I said its beta relative to the portfolio ie cov(asset, portfolio) not cov(asset,index) like usual. Solve such that w x beta is equal for every constituent such that sum w = unity.

I'm using newtons method with interface in Excel and Matabele doing the heavy lifting behind that.

Please don't think I think I'm preaching the gospel here I just tried something out and thought I'd ask what the big shots here do.

I experiment with different portfolio allocation ideas to try to get to something that is intuitive and easy long run.

Mostly I believe in the 70/30 portfolio except varying the allocation from time to time depending on market conditions.
I didn't put any definition of risk parity here. I tried to understand what you posted.

Risk parity would not come up with negative weights unless the beta or the covariance was negative. See the example in the link: http://people.umass.edu/~kazemi/An%20In ... Parity.pdf.
Newbie
Jul 6, 2016
37 posts
19 upvotes
I personally use a risk parity approach for my parents RRSP account. Historically it gets the same return as a balanced portfolio but with lower volatility but with the negative interest rate environment it is quite lively there would be unexpected inflation in medium long term. So a slightly higher allocation to commodities (not gold) and stocks seems logical.
Deal Addict
Apr 22, 2014
3097 posts
474 upvotes
Oshawa, ON
Ironcat wrote: I didn't put any definition of risk parity here. I tried to understand what you posted.

Risk parity would not come up with negative weights unless the beta or the covariance was negative. See the example in the link: http://people.umass.edu/~kazemi/An%20In ... Parity.pdf.
I don't see how negatives would come under question. Especially in a 3 asset universe.
Deal Fanatic
User avatar
Jun 11, 2001
9475 posts
1610 upvotes
Sunlife Granite 2045 AGG fund, lol, 0.71 MER through company plan so it isn't horrible. Allocation is about currently 80/20 (equities/bonds~fixed).

Set and forgot, haha.
...zzz...zzz...zzz...

www.heatware.com
Deal Addict
Apr 22, 2014
3097 posts
474 upvotes
Oshawa, ON
teckie99 wrote: I personally use a risk parity approach for my parents RRSP account. Historically it gets the same return as a balanced portfolio but with lower volatility but with the negative interest rate environment it is quite lively there would be unexpected inflation in medium long term. So a slightly higher allocation to commodities (not gold) and stocks seems logical.
How do you determine allocations? Do you have an inclusion/exclusion criteria? What's your asset universe?

Top