Standard deviation vs VAR

[MrSmart]

@S666

@MrSmart

@S666

In theory to calculate an analytical VaR you don’t actually need any money manager performance history, all you need is price history for the assets currently held in the portfolio. Analytical VaR is based on historical prices of individual securities held in a portfolio at a point in time. On day one of a new strategy, if the manager buys some equities and holds them for a second, a parametric/analytical VaR can be easily calculated, whereas the standard deviation of money manager returns can not as there is no manager performance history. In terms of VaR, only an historic VaR requires a money manager track record to exist before being calculated. Even a Monte Carlo simulated VaR can be calculated without a money manager track record.

So this is surely why VaR is more useful that standard deviation when assessing new money managers.

The same exact methods can also apply to standard deviation, VaR just quantifies the risk in terms of dollars.

The answer given in the OP “Answer: standard deviation typically requires several years before the managers return history is available, which limits its use in determining the effectiveness of new managers and strategies.” states they are concerning themselves with the standard deviation of the managers returns, not the standard deviation of the portfolio holdings - two very different things!

True, but the same limitations apply to VaR in this case. I think the answer is tricky in itself, both are not a good measure. But saying standard deviation is better is still correctly false. If that makes sense.

[S666]

@MrSmart

@S666

@MrSmart

@S666

In theory to calculate an analytical VaR you don’t actually need any money manager performance history, all you need is price history for the assets currently held in the portfolio. Analytical VaR is based on historical prices of individual securities held in a portfolio at a point in time. On day one of a new strategy, if the manager buys some equities and holds them for a second, a parametric/analytical VaR can be easily calculated, whereas the standard deviation of money manager returns can not as there is no manager performance history. In terms of VaR, only an historic VaR requires a money manager track record to exist before being calculated. Even a Monte Carlo simulated VaR can be calculated without a money manager track record.

So this is surely why VaR is more useful that standard deviation when assessing new money managers.

The same exact methods can also apply to standard deviation, VaR just quantifies the risk in terms of dollars.

The answer given in the OP “Answer: standard deviation typically requires several years before the managers return history is available, which limits its use in determining the effectiveness of new managers and strategies.” states they are concerning themselves with the standard deviation of the managers returns, not the standard deviation of the portfolio holdings - two very different things!

True, but the same limitations apply to VaR in this case. I think the answer is tricky in itself, both are not a good measure. But saying standard deviation is better is still correctly false. If that makes sense.

Totally agree it’s an awful, awful question and definitely wouldn’t be found in a real CFAI exam no doubt. I was just trying to get the best answer to an awful question.

But I disagree that the same limitations apply to VaR - read my previous post. You can calculate a VaR on day 1 of a new portfolio - you can’t calculate the standard deviation of returns on day one of a new portfolio. That is just explicitly true.

[MrSmart]

@S666

Totally agree it’s an awful, awful question and definitely wouldn’t be found in a real CFAI exam no doubt. I was just trying to get the best answer to an awful question.

But I disagree that the same limitations apply to VaR - read my previous post. You can calculate a VaR on day 1 of a new portfolio - you can’t calculate the standard deviation of returns on day one of a new portfolio. That is just explicitly true.

If you mean day 1 of the new managers portfolio performance, you can for both. But the sample is too small to be statistically significant.

[S666]

No - I think you are misunderstanding what analytical/parametric VaR is and how it is calculated. It is calculated using the historic standard deviations/returns/covariances and correlations between the assets held in the portfolio AT A POINT IN TIME.

If you are talking about historic VaR then I agree with you. But I am not - I am talking about analytical/parametric VaR

All you need are price histories for the assets held in your portfolio, and you can calculate your VaR - there is no issue of “small sample size” as you are using full, real price histories of the assets held in your portfolio.

On day 1, you can get a full VaR reading back for any time frame for which you have individual asset price histories. Again, there is absolutely no small sample size issue with parametric VaR.

[MrSmart]

@S666

No - I think you are misunderstanding what analytical/parametric VaR is and how it is calculated. It is calculated using the historic standard deviations/returns/covariances and correlations between the assets held in the portfolio AT A POINT IN TIME.

If you are talking about historic VaR then I agree with you. But I am not - I am talking about analytical/parametric VaR

All you need are price histories for the assets held in your portfolio, and you can calculate your VaR - there is no issue of “small sample size” as you are using full, real price histories of the assets held in your portfolio.

On day 1, you can get a full VaR reading back for any time frame for which you have individual asset price histories. Again, there is absolutely no small sample size issue with parametric VaR.

Oh, I thought you meant the price performance and correlations for that holding period only. In this case, the manager can be heavily active in management, so it’s pointless either way. Remember that we’re talking about the manager’s performance, and not the portfolio risk at a point in time.

On topic, you could still calculate the SD of the portfolio held at that point in time, using the historic standard deviations. That doesn’t provide an advantage over normal conditions.