Algo Selection at Quantopian

By Dr. Jess Stauth*, Quantopian’s Managing Director on the Investment Team, transcribed by HFI


Quantopian is a Boston-based company that has created a crowd-sourced hedge fund by facilitating freelance quantitative analysts to develop and test trading algorithms. The firm licenses the best algorithms to buy and sell securities. Quantopian is backed by venture capital firm Andreessen Horowitz, and $2m from the venture capital arm of Steve Cohen’s Point72 Asset Management. Point72 has also committed to allocate up to $250m of capital to invest in the funds developed. Quantopian’s investment process can be divided into four parts depicted below:

quantopian inv process 2

This webinar is focused on the selection of algorithms.

quantopian algorithm-evaluation 2


The Author Diligence part of the selection process (part 3 above) is started with an e-mail form laying out the Quantopian process in detail and asking for responses to a series of basic questions about the author. This is followed by a Skype call or telephone call. In-coming algorithms are time-stamped and stored with version control in the Quantopian database.

In the due diligence questionnaire sent to authors the economic rationale for the algorithm has to be explained and that explanation is cross-checked with the algorithm/dataset for consistency and logic.

In operational due diligence Quantopian does not alter the code of the algorithm under consideration – that is fixed. Rather such elements as the APIs or changes to an index construction or model are reviewed and brought up to date.

Algorithms that have passed through due diligence go to the monthly formal meeting in which the CIO of Quantopian makes a decision on inclusion of the submitted algorithm to the production portfolio.


Common Failings of Submitted Algos

  1. Out of sample performance does not match in-sample performance – commonly because of over-fitting.
  2. Returns exhibit factor risk such as to markets (e.g. correlation to S&P500 Index), sectors or to the Dollar. As these factors are not diversified away through portfolio construction they must be tightly controlled at the algorithm level. There are 11 sectors controlled for and 5 risk factors.
  3. Algorithms that combine investment styles are difficult to evaluate without more context. Quantopian might work with the authors to separate the components and evaluate them apart.
  4. There is a bias against algorithms that produce returns that show high Fama-French factor contributions like momentum or smaller market cap.


Quantopian is looking for:

  1. Algos with low common factor risk by design wherein returns are driven by stock selection.
  2. Consistency in the size of exposure at  the (algorithm) portfolio and position levels.
  3. Consistency in the pattern of return, avoiding strategies for which the overall return depends on a few hot-spots of high positive return.
  4. Algorithms that work within the 1500 stock universe of trade-able names specified by Quantopian (stock list available as a single line API import from the firm).


Points from the webinar Q&A

  • Quantopian does not mind a strategy that requires leverage to implement.
  • Quantopian have one vehicle at the moment – a pure alpha portfolio, launched in June 2017 for institutional investors.
  • Weightings in the traded portfolio of Quantopian are a function of (equal) risk weighting, volatility (inverse relationship), size of beta, liquidity and scalability, and pure alpha. Quantopian runs a monthly portfolio rebalance process.
  • Quantopian will look at strategies run on external data sources rather than the data available at the Quantopian store, but that does make it a two-stage process for the firm, with all the implications of that.
  • There were 14 algos trading live in September.
  • The fewer free parameters in the algorithm the better is a general bias.
  • A development to come shortly will be the ability to look at strategies that use futures or futures in combination with equities.
  • Authors of algorithms sign an exclusive three-year license with Quantopian. There is a break clause after a year. Profits due to authors of algorithms are accumulated daily and paid out annually.



*extracted from the webinar, “Making the Grade” by Dr. Jess Stauth, from the week of the 11th September 2017.