Dear Community,

In a very hard decision, we've elected to cease support for the current version of Alpha Streams(v1.0) and focus our efforts on a new version that would implement a few core quant principles at its core (2.0).

The current submission and filtering process seeks strategies that perform well in all market regimes. This is a relatively unrealistic task and results in strong overfitting. A much more realistic approach is finding small factors which apply sometimes and swapping those factors in and out as regimes change with a global-macro/portfolio-level intelligence.

Additionally, providing too much information about the asset always led to selection bias. The new Alpha Streams universe will be a fixed portfolio of 5,000-6000 assets, anonymized, and scaled in value to 1.0 to help avoid this bias.

Keeping crowd-alpha accessible to investors is important to our mission, so we will build a way to bundle these factors for individual portfolios. Alpha Streams 1.0 only supported one alpha per live server, but there should be thousands or millions of these factors, each much smaller and focused on a specific idea. Ideally, we should support running hundreds of these factors from a single live-host server. 

What does this mean for current investors?

  • Alpha renewals have been disabled, and we will work with existing longer-term licensing clients to find a solution for their alphas or refund the license.
  • Stop any active live trading servers as they will automatically stop in the next few days. Ensure you tidy up any remaining holdings. 
  • Investors passionate about their current alpha reach out to support@quantconnect.com and we'll assist to arrange a lifetime license to the alpha if the developer is open to it.


What does this mean for authors/engineers?
We're incredibly grateful for those who've worked with QC to make Alpha Streams so far. We hope the reasoning for why we're doing this shines through and you agree it is the best for the investors long term. It will take some time but please bear with us while we refactor the platform.

  • All existing license obligations are “relinquished”; we'll contact you if an investor would like a lifetime license.
  • We'll work to pay out all existing wallet balances for alpha licenses.   
  • Alpha live hosts will be stopped in the next week. 
     

Why not just run 1.0 and 2.0 in parallel?

We see some of these issues as too fundamental to keep the system public. In the interest of wanting the best for the investors using QC, keeping their faith in the long-term, and focusing our efforts, we elected for a full stop. The ultimate result of the overfitting is generally poor performance out-of-sample which makes it hard for us to promote in good conscience. 

O8G-XgfewhTXXthuvYAQeCsDRgDm3pqidjgmEet8AnLwoEPYcCI_dV6yTb2sIy8LFiX6R8UrAMyRmw_tOjZ56kdTyLZZR-oO_HS6R4WbvSDdoY8_vWxb6O2eLHz2hSh9K92D990R

Negative Skew of Alpha Sharpe Ratios vs S&P500 or Russel Constituents

677_1648602533.jpg

Negative Skew of Alpha Sharpe Ratios vs S&P500 or Russel Constituents (6mo)

To measure the performance we compared distributions of the Alphas Sharpe Ratio with S&P500 and Russel constituents. As seen above, there was a population shift negative(blue=alphas) which made constructing a portfolio of alphas difficult. We researched taking a “needle in a haystack” approach and only selecting the top 5% of the Alpha Market but after eliminating illiquid alphas, and a few crypto outliers, the remaining alphas underperformed the S&P500. We also explored taking uncorrelated alphas and adding them to a broad market portfolio to complement performance but they were not additive. 

I know it is tough news for many who've invested months in designing and submitting Alphas. It will take some time but we will carry those skills over into the new Alpha 2.0 platform.  The new version will adapt lessons from the industry's titans such as Rentech who've used this style of investing successfully for 30 years. We will be the first to bring this to the wider investor market with the help of an awesome community to create the millions of factors needed.  This technology could be used for QC Alpha Streams or independent quant firms building on QC infrastructure. Feel free to ask any questions here and we'll do our best to answer them,

Best,

Jared 

Author