mikko, from what i understand, after you pass the first two conditions (profit and low beta to spy), the rest of the criteria are equal weighted and scored relative to the rest of the field.

I would like to have a comment directly from Q staff as I can adjust profits(and sharpe/sortino) by leverage but as everyone knows the DD/downside vola then increases so I would like to know what are acceptable levels.

IMHO it would be quite wise for Q to normalize curves to certain leverage but as they don't do this and profit is quite important it all comes to adjusting leverage as high as possible so that downside volatility/DD is controlled to some level. The question then comes what is "low DD" mentioned in the contest page. Adjusting to high sharpe or sortino is easy as it's numeric value.

I think they're keeping the weights a bit vague so people don't try to game the system.

Hi Mikko,

Each category is equally weighted. The score is a relative score based on the competition in the current contest. You are essentially ranked on your percentile rank for each category, evenly weighted. Of course, as Toan mentioned, getting all three badges is the first priority, any algo with all three badges will be ranked higher than any algo with 2 or fewer.

Thanks Jamie! Exactly the information I wanted - the score is based on ranks and is equally weighted.

One question about beta, my algo has ~10 year backtest beta of almost zero (-0.03) but at some periods it may be higher, last 2 years backtest it was 0.3. Would this disqualify my algo from beta badge (I can adjust the beta if that is needed by adjusting my hedging ratio but that will affect the returns negatively)? I'm in the game for long term and at certain periods (uptrending) the beta naturally just is higher than in some other periods (downtrening).

Thanks!

Hi Mikko,

To see the details on the beta calculation, check out the rules page and scroll to the "Judging" section. The result of this calculation will have to be less than 0.3 in the 2-year backtest.

Hello,

I have now entered to contest with one of my algos. My beta since 2007 (this is the first date of some data I'm using) is -0.05. However last 2 years my beta has been 0.33. This is due to market trending up so it's only reasonable that the beta is not negative as the algo is also making profit on the same period.

Now the question is, is the 2 year beta test really reasonable or a good predictor of future risk for algos that are meant to work for longer term? As I said, my longer term beta is -0.05. note how this particular algo works at 2007-2009 period when sp500 was having massive drawdown and see the beta there.

I assume algos are meant to work in different kinds of market regimens, isn't the long term beta exactly what should be measured, not the short term one?

I would like to hear everyones thoughts on this.

Attached are 3 backtests of the algo I'm talking about (these are only screenshots - the algo is private)

Here is a test period including both downtrend (2007-2009) and uptrend (2009-). Note how beta is negative as it should be at downtrend.