What are the future research topics in this field?Thanks a lot in advance.M.Likelihood Serving the Quantitative Finance Community
-
- MLikelihood
- Posts: 0
- Joined:
About portfolio costruction optimization
Hallo,I'm interested to have an overview on portfolio costruction's methods/theories. I'm also interested to learn something about the most powerful software for portfolio.Expecially I wuold to know:1) How many portfolio's models/methods are there? I know only Markovitz family models (var/covar, VaR, maxDD) and CAPM model.2) What are the most used by financial istitutions like asset manager, hedge fund etc...? So, what's the industry standard?3) What are the best software house for portfolio's applications? I know only Alternativesoft, Finanalytica and Axioma.4) Is there a website with resources about portfolio like riskmetrics for VaR?5) Finnally, What are the future research topics in this field?Thanks a lot in advance.M.Likelihood
What are the future research topics in this field?Thanks a lot in advance.M.Likelihood -
- MLikelihood
- Posts: 0
- Joined:
About portfolio costruction optimization
Hi,as accademic student, so without any practical experience, I think that a theory made in '50 (MPT) is to much old for modern total return investment management. Isn't it?Tnx a lot.M.L.
About portfolio costruction optimization
I suggest you pick up Frank Fabozzi's "Financial Modeling of the Equity Market: From CAPM to Cointegration", Wiley Finance.
About portfolio costruction optimization
QuoteOriginally posted by: MLikelihoodHallo,Expecially I wuold to know:1) How many portfolio's models/methods are there? I know only Markovitz family models (var/covar, VaR, maxDD) and CAPM model.2) What are the most used by financial istitutions like asset manager, hedge fund etc...? So, what's the industry standard?3) What are the best software house for portfolio's applications? I know only Alternativesoft, Finanalytica and Axioma.4) Is there a website with resources about portfolio like riskmetrics for VaR?5) Finnally, What are the future research topics in this field?Thanks a lot in advance.M.Likelihood(1) Academic models are different from practical ones for a few reasons: (i) in academics, usually it is assume that the list of stocks is given & the optimization only needs to solve for the weights, however in the practical we need to determine both the list of stocks and the weights, i.e. from a universe of 1000 stocks pick both the best 10 stocks (based on some alpha/return expectation) and the corresponding weights.(ii) in practical world, we have additional constraints that either the weights are a certain minimum or zero, i.e suppose we want a portfolio of 10 stocks where we want each stock with a minimum weight of 5% and we have a universe of 1000 stocks - and the constraint will be for each of the 1000 stocks, either the weight will be zero (i.e. the stock is not included) or the weight is greater than 5% - i.e. (w=0 Or w>0.05) - i.e. the constraint will be "w*(w>0.05)=0" - such constraints are hard to solve using academic methods. In practice, the optimizers use an "iteration+optimization" approach. This is a fairly complex issue - more details are probably beyond the scope of this forum & often proprietary.(iii) A third issue multi-period nature of practical optimizers versus single-period academic ones. One the issue is the turnover. Many optimizers allow the turnover as a part of either the constraints or the objective. Basically, the idea is that on every re-balance date, one may want to restrict the new portfolio to deviate only a fraction (such as say 20%) from the old portfolio. This is different from just using the transaction cost in the objective function since the transaction cost often can include stock borrowing and stock lending costs for the entire portfolio rather than the part of the portfolio that is just being re-balanced.(2) There is no industry standard as such. A lot depends on the capability and the kinds of constraints that the optimizers can handle. For instance, there still isn't any optimizer that can natively construct a market-neutral portfolio. Many optimizers also have limitations on the risk models that can be used. Some are generic and others only allow a specific risk model (usually of the same vendor). Most optimizers can either solve for "Max Returns - c1 * Risk from Risk Model - c2 * Transaction Cost Model - c3 * Deviations from Benchmark" Or solve for "Max Returns - c11* Specific Risk from Risk Model - c12*Residual Risk from Risk Model - c2* Transaction Cost Model - c3 * Deviations from Benchmark" where c1, c2, c3, c11, c12 are Risk Premium & Others Premiums specified by the portfolio manager.(3) Northfield and Barra optimizers are probably quite popular. Most firms tend to buy a commercial optimizers, but a smaller number particularly those who use serious quantitative stock picking often use basic optimization capabilities in softwares such as Matlab etc and wrap it with various iterative shell to take care of practical constraints.(4) Riskmetrics doesn't fit into the practical portfolio construction framework. The leading industry standard Risk model is the Barra Risk Model, followed by a distant second which is the Northfield Risk Model(5) Too general question. Be more specific.Cheers
What are the future research topics in this field?Thanks a lot in advance.M.Likelihood(1) Academic models are different from practical ones for a few reasons: (i) in academics, usually it is assume that the list of stocks is given & the optimization only needs to solve for the weights, however in the practical we need to determine both the list of stocks and the weights, i.e. from a universe of 1000 stocks pick both the best 10 stocks (based on some alpha/return expectation) and the corresponding weights.(ii) in practical world, we have additional constraints that either the weights are a certain minimum or zero, i.e suppose we want a portfolio of 10 stocks where we want each stock with a minimum weight of 5% and we have a universe of 1000 stocks - and the constraint will be for each of the 1000 stocks, either the weight will be zero (i.e. the stock is not included) or the weight is greater than 5% - i.e. (w=0 Or w>0.05) - i.e. the constraint will be "w*(w>0.05)=0" - such constraints are hard to solve using academic methods. In practice, the optimizers use an "iteration+optimization" approach. This is a fairly complex issue - more details are probably beyond the scope of this forum & often proprietary.(iii) A third issue multi-period nature of practical optimizers versus single-period academic ones. One the issue is the turnover. Many optimizers allow the turnover as a part of either the constraints or the objective. Basically, the idea is that on every re-balance date, one may want to restrict the new portfolio to deviate only a fraction (such as say 20%) from the old portfolio. This is different from just using the transaction cost in the objective function since the transaction cost often can include stock borrowing and stock lending costs for the entire portfolio rather than the part of the portfolio that is just being re-balanced.(2) There is no industry standard as such. A lot depends on the capability and the kinds of constraints that the optimizers can handle. For instance, there still isn't any optimizer that can natively construct a market-neutral portfolio. Many optimizers also have limitations on the risk models that can be used. Some are generic and others only allow a specific risk model (usually of the same vendor). Most optimizers can either solve for "Max Returns - c1 * Risk from Risk Model - c2 * Transaction Cost Model - c3 * Deviations from Benchmark" Or solve for "Max Returns - c11* Specific Risk from Risk Model - c12*Residual Risk from Risk Model - c2* Transaction Cost Model - c3 * Deviations from Benchmark" where c1, c2, c3, c11, c12 are Risk Premium & Others Premiums specified by the portfolio manager.(3) Northfield and Barra optimizers are probably quite popular. Most firms tend to buy a commercial optimizers, but a smaller number particularly those who use serious quantitative stock picking often use basic optimization capabilities in softwares such as Matlab etc and wrap it with various iterative shell to take care of practical constraints.(4) Riskmetrics doesn't fit into the practical portfolio construction framework. The leading industry standard Risk model is the Barra Risk Model, followed by a distant second which is the Northfield Risk Model(5) Too general question. Be more specific.Cheers
Last edited by pb273 on March 1st, 2007, 11:00 pm, edited 1 time in total.