Back to Community
Third-Party Challenge questions/comments

The most recent challenge ($10k-third-party-challenge-design-a-factor-for-a-large-us-corporate-pension) is not entirely clear to me:

"The portfolio must hold at least 100 assets in the QTU"

Does this mean that assets not in the QTU can be traded (e.g. ETFs)? Or that the portfolio is constrained to the QTU, and must hold at least 100 assets? Presumably, assets must come from the QTU, but it isn't stated explicitly.

Is the idea that leverage should be ~1.0, or are we also looking for predictions of when it should be lower or higher? Would going to all cash be allowed, for example?

Generally, why not just stick to a ranking across the entire QTU end-of-day? We are basically saying that we want a prediction of not to hold certain stocks, right? Is this what the customer needs? Is the customer wanting us to say that no prediction is possible for certain assets at a given time, or we are to kinda slice off the cream of the crop, even though there may be some residual predictability across the QTU? The prior wise guidance, I thought, was that ranking across the entire QTU was desirable (and not to use the optimizer, which tends to lead to less dispersion in ranks).

"There are no constraints on risk exposures or beta to SPY, but your exposures must be time-varying — these tilts should be moving daily."

This requirement sounds like one could deviate from a market-neutral portfolio (e.g. long-only or short-only)? Or maybe we don't care and are looking just for predictability? Also, what is meant by having exposures time-varying and moving daily? If this is actually a requirement, then how should we measure it and what are the pass/fail criteria?

"The specific Sharpe ratio over the first 5 days must be positive."

When I first read this, I thought it was referring to the first 5 days of the backtest. But then I realized it might refer to the IR versus delay plot in the tearsheet. What does this requirement mean?

10 responses

Another question I have is whether sharing the code one's challenge algo would make one ineligible for the challenge? For an allocation from the third party? I prefer to share my code publicly, but maybe this practice has implications in this context?

"There should be a daily turnover of 5% to 20%."

How are we to determine pass-fail for turnover?

Assuming long-only and short-only would be o.k., would it be preferred to separate algos into long-only and short-only?

I'd be interested if someone could explain how the uniqueness score is being computed, and if it could be implemented in an algo (e.g. as an optimization constraint). I took a quick look at the corporate_pension_challenge_jan_2020.csv DataFrame and there were multiple columns by date; I'd expected a single column of returns. What are the columns and how do they allow a correlation check with the alpha factor returns data if they are somehow obfuscated?

First off, thank you for all the questions! We appreciate your deep engagement with the post.

One thing to reiterate – it's important to remember that we will implement a strategy in cohesion with many other strategies. That means that many of these rules have "soft bounds" - your colleagues might have come up with a strategy that naturally hedges or balances out with yours. We provide these purely as guideposts to give you some target ranges, and as you move outside them, we have to look harder to find offsetting portfolios to balance them out. If you have a 21% turnover portfolio, we won't really be concerned. If you have a 50% turnover, you're pretty far outside the guidepost.

We typically drop assets outside the QTU during implementation as they tend to fail one of our constraints: insufficient liquidity, market cap thresholds, hard to borrow, MLPs, etc. ETFs are fine. We look at your full portfolio and then just the QTU portion, so if you are long QTU/short non-QTU, it will put you at a disadvantage as we'll identify 50% of your portfolio as "difficult to implement."

You should focus on where you think your advantage is – if you feel you have a good ability to time the market we will analyze it such that we look at your ability to vary your bet direction and size. Do you go from +10% to -10% and from +10% to +20% effectively? Or are you always +10% and when you go to +20% those periods are relatively underperform? If you believe you have better ability to create a market neutral portfolio, time trying to "time leverage" is probably better spent refining your idea. Remember, our goal is to keep the portfolio invested.

You can deviate from a market neutral portfolio – there is no hard and fast "fail" threshold. However, your average beta-to-SPY over a < 4-week period should probably average out to 0. Ranking a broader universe of stocks is always preferred, but there are often sector and industry specific insights that are on a smaller universe of names that remain interesting.

I don't want to say too much about how the uniqueness score is computed, and you shouldn't use it in the optimizer (or use the optimizer in the first place).


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.

Thanks Thomas -

Some follow up:

ETFs are fine.

What does this mean? So, could the algo trade only ETFs, and then you would determine the stocks in the ETFs to see that the requirement of "at least 100 assets in the QTU" is met? Or would one have to trade at least 100 stocks in the QTU, and then on top of that, some ETFs could be traded? Or did you mean something else?

I don't want to say too much about how the uniqueness score is computed, and you shouldn't use it in the optimizer

I guess you have your reasons, but I'd think it would be the perfect thing to use in an objective function for portfolio optimization. What's the point in doing it manually; that's what we have computers for, right?

You can deviate from a market neutral portfolio – there is no hard and fast "fail" threshold. However, your average beta-to-SPY over a < 4-week period should probably average out to 0.

Well, 4 weeks isn't very you are saying you want market-neutral as a requirement, as I see it. I suggest you add this to your post.

What about dollar neutral? Long-only? Short-only? Anything in between? I think you are saying that dollar neutral is not a requirement.

One suggestion would be to use the requirements you've already laid out on, and then describe which of those pertain as-is, which ones are different for this challenge, and which additional requirements have been added. Putting things in a table is sometimes helpful.

we will implement a strategy in cohesion with many other strategies

How are you combining strategies? Specifically, I'm wondering if you use directional (long/short) information, or if you convert to unsigned scores? For example, say one submitted a short-only strategy. Would it then need to be combined with a long-only strategy to achieve dollar neutrality?

Or are you using the end-of-day positions as alpha factor scores, per the definition on :

An alpha is an expression, applied to the cross-section of your universe of stocks, which returns a vector of real numbers where these values are predictive of the relative magnitude of future returns.

In other words, you throw out the directional information, and are just interested in the relative scores across the universe.

Is there any relevance that the source of capital is "a large US corporate pension fund" or is this just promotional information?

I find it a bit surprising that the pension fund would be interested in excluding exactly the same common returns factors as Quantopian? Please explain.

I gather that Quantopian will not be offering algos for direct investment by the pension fund, but will somehow be combining them, and potentially offering a single investment. It would be interesting if you could elaborate on what you intend to offer.

Any idea if ETFs are allowed in the Third-Party Challenge?

I’ve confirmed that ETFs are ok but one still needs to hold at least 100 stocks in the QTU. Does anyone understand why the pension fund would care about having a certain number of stocks, and from the QTU?

Maybe someone from the pension fund can speak to this?