Strange contest scoring?

Hi!

Do you also think, that algorithm ranking/evaluation is very strange? There's to big emphasis on drawdowns, volatility and stability, while returns are not very important. That way the contests are winning algorithms, which are very stable but profits are also very low. After all, which algorithm would you prefer?

ALGO1: 40% annual returns, 7% annual volatility and 3% max drawdown
or
ALGO2: 5% annual returns, 2% annual volatility and 1% max drawdown

What do you think?

10 responses

I think that the answer to your question depends on what you're looking for in an algorithm. I imagine that a crucial point for Quantopian is that they want to be confident that they know how to estimate when past performance will continue into the future, and I imagine that lower volatility helps with that.

I understand, but i still think, that the meaning of volatility is emphasized to much. In example above: for 3.5x worse volatility you get 8x better returns.
7% volatility represents only 17.5% of returns (if returns are 40%), but 2% volatility represents 40% of returns (if returns are only 5%). That means, that you are still risking much larger part of your profits. And still ... ALGO2 got approx. 30 points more than ALGO1. (contest)

Under the line ... if you take in account risk(volatility) and reward (returns), with ALGO1 the probability of bigger returns is much higher than with ALGO2.

I think, that this is happening, because Quantopian has more indicators measuring risk than indicators measuring reward, but they are all weighted equally...

Gregor, it's all about what one wants from the contest. It's very reasonable for someone to look at the two examples you created and say "hey, I like algo 1, but Quantopian likes algo 2!" The difference, to state the obvious, is that we want different things.

The contest is designed to encourage algorithms that fit our allocation profile. So why does Quantopian want these low-volatility algorithms for our allocations? Our investment clients are in the market to make investments that can be levered up 4-6X. With access to that kind of leverage, a low-return algorithm can be very interesting - so long as it has low volatility.

Hopefully that helps explain why we're looking for algorithms that don't match what you would have expected.

All that said, we're working on a revision to the contest that will change the way we all think about the problem. The key outcomes, the low-volatility algorithms, aren't going to change, but the way we evaluate and score them will be very different.

Disclaimer

The material on this website is provided for informational purposes only and does not constitute an offer to sell, a solicitation to buy, or a recommendation or endorsement for any security or strategy, nor does it constitute an offer to provide investment advisory services by Quantopian. In addition, the material offers no opinion with respect to the suitability of any security or specific investment. No information contained herein should be regarded as a suggestion to engage in or refrain from any investment-related course of action as none of Quantopian nor any of its affiliates is undertaking to provide investment advice, act as an adviser to any plan or entity subject to the Employee Retirement Income Security Act of 1974, as amended, individual retirement account or individual retirement annuity, or give advice in a fiduciary capacity with respect to the materials presented herein. If you are an individual retirement or other investor, contact your financial advisor or other fiduciary unrelated to Quantopian about whether any given investment idea, strategy, product or service described herein may be appropriate for your circumstances. All investments involve risk, including loss of principal. Quantopian makes no guarantees as to the accuracy or completeness of the views expressed in the website. The views are subject to change, and may have become unreliable for various reasons, including changes in market conditions or economic circumstances.

OK, let's assume, that ALGO2 is leveraged 5x and ALGO1 leverage stays 1x, now let's compare the data...

ALGO1: 40% annual returns, 7% annual volatility and 3% max drawdown
ALGO2 (leveraged 5x): 25% annual returns, 10% annual volatility and 10% max drawdown

I understand, that if both algorithms are leveraged is better to have lower volatility and drawdown, but ALGO1 is still performing much better than ALGO2 with leverage 5x. And if we consider, that leverage can have "strange impact" on returns, risk with leveraged ALGO2 are much higher...

Again, am i missing something?

Hi gregor,

Are you familliar with MPT?

Your algorithms sharpe ratios are wildly different, so in the contest of your question yes, algo 1 would always be chosen. But you are missing the point I think, because the ranking system would not favor the algo2 as you think it would.

What would be a relevant question is :given two algorithm with identical sharpe ratios, Q would favor the one with the lowest possible volatility. Is it fair?

In fact, we know that choosing the least risky algorirhm is a risk aversion preference.

The fairness argument in turn, would spring from the fact that any algorithm could be adjusted to any risk preference with the addition of a risk free asset.

Since Q risk aversion is strong, people are submitting algorithm skewed toward that risk preference by allocating little capital.

But an asset manager cannot let most of its money stay in the risk free asset for long can he ? Clients, whatever their risk preference, would be annoyed by that.

The risk aversion argument aside, all agree that the ranking is indeed in need of improvement

Hey Q, Better sooner than later right :-)?

Hey @Dan, could we please extend the backtester to add the stability metric (that is present in the contest but not the backtest). Could we also have the backtest volatility metric to be shown to 3 or 4 decimal places so that the number can provide better feedback to the algo developer on whether a change improved the algorithm in that regard (as of now it shows one decimal place like 0.03 or 0.04 which is quite unhelpful in a metric that Q attaches so much importance to. 0.0314 or 0.0435 are more informative likewise 3.14% or 4.35%). Backtest Alpha display would benefit from the same change. It would also be nice if a score like the contest score 0- 100 can be displayed in a backtest that shows how close the backtest is with regard to the allocation profile. With the 3 algo limit in the contest, we have no feedback on non-contest algos. Waiting 6 months in a vaccum without any feedback can be unproductive, an automated score is our best chance to guage what to improve upon. Thanks.

Hi Leo,
Yes, I second that! On all the points you mention.
Cheers.

https://ibb.co/ik8a2b

Interesting fact (image), algo2 is nmb1 in contest and even with leverage quite useless for the investors... What do you think?
True, algo1 has bad beta, but this will even out when the bear market comes...