Back to Community
For Hedge Funds: Accern Alpha Stream News and Blog Algorithms and Data Samples

Dear Community Members,

I am dedicating this thread specifically for hedge fund managers who are interested in Alpha Stream, Accern's institutional news and blog data product. I want to have a central location to preserve all our current and future algorithms and sample data to showcase to hedge funds in a seamless manner. This thread will be updated weekly or bi-weekly with new algorithms and data set to be demonstrated. Community members who are interested in pursuing a career in the hedge fund space will also benefit from this thread as well.

About Accern

Accern is the world’s first big data media analytics provider to deliver the most comprehensive dataset of actionable and authentic stories and analytics from over 20 million news and blog sources for quantitative trading. With cutting-edge machine learning, deep learning, and neural network algorithms, we have designed actionable trading metrics for the quantitative trading space. We currently utilized the data as a standalone model, but it can be applied very successfully with any multi-factor models.

Alpha Stream Trading Metrics

Sentiment Analysis

  • Article Sentiment (1 to -1): Identifies the attitude the article is written in. This can be used as a directional signal.
  • Story Sentiment (1 to -1): Tracks the aggregated sentiment for a specific story. This can be used as a directional signal.
  • Average Day Sentiment (1 to -1): Aggregates article sentiment for a company each day. This can be used as a directional signal.

Rankings

  • Overall Source Rank (1 - 10): Evaluates the credibility of a source based on its timeliness and re-post rate of releasing stories.
  • Event Source Rank (1-10): Evaluates the credibility of a source based on its timeliness and re-post rate of releasing stories on specific events.
  • Overall Author Rank (1 - 10): Evaluates the credibility of an author based on its timeliness and re-post rate of releasing stories.
  • Event Author Rank (1-10): Evaluates the credibility of an author based on its timeliness and re-post rate of releasing stories on specific events.

Impact Analysis

  • Overall Event Impact Score (1-100): Probability that an event will have a greater-than-1% impact on any stock.
  • Entity Event Impact Score (1-100): Probability that an event will have a greater-than-1% impact on the mentioned stock.

Time and Exposure Analysis

  • First Mention (TRUE/FALSE): Alerts you on unique story before they become exposed on the web.
  • Story Saturation (Low/Mid/High): Tracks the online exposure rate of a specific story. This can be used as an enter and/or exit signal.

Hedge Fund S&P 500 Strategy A: Long / Short (Weighted)

We have conducted a backtest using a Long/Short strategy with weights. This is a weekly holding period strategy. Every Monday at 9:30 AM when the market opens, we would identify stocks that matches the criteria for our “bull” basket and enter those “bull” stocks into long positions. A stock will match our “bull” criteria if a trustworthy source released a positively-toned story with high probability of impacting its price. We will also identify stocks that matches the criteria for our “bear” basket and enter those “bear” stocks into short positions. A stock will match our “bear” criteria if a trustworthy source released a negatively-toned story with high probability of impacting its price. All positions are closed on Friday 3:45 PM.

ADDING WEIGHTS: We add a twist to this strategy by adding weights to some metrics. For example, the more positive the sentiment and higher the impact of an article that mentions a company, the more shares of that company we will buy. The more negative the sentiment and higher the impact of an article that mentions a company, the more shares of that company we will short.

NOTE: This backtest will take ~10 minutes to start due to the massive 5 million article history.

BACKTEST REPORT (PDF)
https://dl.dropboxusercontent.com/u/428478238/AlphaStream%20Backtest%20Report%20(LongShort).pdf

S&P 500 – Full Historical Data

1 CSV with over 5 million articles
https://www.dropbox.com/s/a7u4hkgeaep0l26/backtest_sp500.csv.gz?dl=0

8 CSV with over 5 million articles segregated
https://www.dropbox.com/sh/6plq6qgfiljmga4/AABxXz-Vv5XqZcRdjIm219_ja?dl=0

S&P 500 – Quantopian Backtest Data (Only 3 Metrics - Small File)

5 CSV with over 5 million articles segregated
https://www.dropbox.com/sh/9eem9dvd2hqmort/AAAg0bA3E3JNX5HNDwATciE1a?dl=0

Enjoy,

Kumesh Aroomoogan
Co-Founder and CEO, Accern
[email protected]

Filtering gif

Clone Algorithm
139
Loading...
Backtest from to with initial capital
Total Returns
--
Alpha
--
Beta
--
Sharpe
--
Sortino
--
Max Drawdown
--
Benchmark Returns
--
Volatility
--
Returns 1 Month 3 Month 6 Month 12 Month
Alpha 1 Month 3 Month 6 Month 12 Month
Beta 1 Month 3 Month 6 Month 12 Month
Sharpe 1 Month 3 Month 6 Month 12 Month
Sortino 1 Month 3 Month 6 Month 12 Month
Volatility 1 Month 3 Month 6 Month 12 Month
Max Drawdown 1 Month 3 Month 6 Month 12 Month
# Backtest ID: 559ebf5a78389f1239b0f626
There was a runtime error.
8 responses

Hedge Fund NASDAQ 100 Strategy A: Long / Short (Weighted)

We have conducted a backtest using a Long/Short strategy with weights. This is a weekly holding period strategy. Every Monday at 9:30 AM when the market opens, we would identify stocks that matches the criteria for our “bull” basket and enter those “bull” stocks into long positions. A stock will match our “bull” criteria if a trustworthy source released a positively-toned story with high probability of impacting its price. We will also identify stocks that matches the criteria for our “bear” basket and enter those “bear” stocks into short positions. A stock will match our “bear” criteria if a trustworthy source released a negatively-toned story with high probability of impacting its price. All positions are closed on Friday 3:45 PM.

ADDING WEIGHTS: We add a twist to this strategy by adding weights to some metrics. For example, the more positive the sentiment and higher the impact of an article that mentions a company, the more shares of that company we will buy. The more negative the sentiment and higher the impact of an article that mentions a company, the more shares of that company we will short.

NOTE: This backtest will take ~5 minutes to start due to the massive 3 million article history.

BACKTEST REPORT (PDF)
https://dl.dropboxusercontent.com/u/428478238/AlphaStream%20Backtest%20Report%20(LongShort).pdf

NASDAQ 100 – Full Historical Data

1 CSV with over 3 million articles
https://www.dropbox.com/s/zq8aoeynqnaw6te/backtest_nasdaq100.csv.gz?dl=0

4 CSV with over 3 million articles segregated
https://www.dropbox.com/sh/54lkseh0aed52eg/AAA4F26vKGk0QkrrHbxuVOhZa?dl=0

NASDAQ 100 – Quantopian Backtest Data (Only 3 Metrics - Small File)

4 CSV with over 3 million articles segregated
https://www.dropbox.com/sh/vxdnrf4d6qx4tz1/AADT5d2QgFBLanokNSVaMaHTa?dl=0

Filtering gif

Clone Algorithm
38
Loading...
Backtest from to with initial capital
Total Returns
--
Alpha
--
Beta
--
Sharpe
--
Sortino
--
Max Drawdown
--
Benchmark Returns
--
Volatility
--
Returns 1 Month 3 Month 6 Month 12 Month
Alpha 1 Month 3 Month 6 Month 12 Month
Beta 1 Month 3 Month 6 Month 12 Month
Sharpe 1 Month 3 Month 6 Month 12 Month
Sortino 1 Month 3 Month 6 Month 12 Month
Volatility 1 Month 3 Month 6 Month 12 Month
Max Drawdown 1 Month 3 Month 6 Month 12 Month
# Backtest ID: 559c192c9dc738124004fb9b
There was a runtime error.

Hedge Fund S&P 500 Strategy A: Monthly Long / Short (Weighted)

We have conducted a backtest using a Long/Short strategy with weights. This is a monthly holding period strategy. At the beginning of each month at 9:45 AM (15 minutes after) when the market opens, we would identify stocks that matches the criteria for our “bull” basket and enter those “bull” stocks into long positions. A stock will match our “bull” criteria if a trustworthy source released a positively-toned story with high probability of impacting its price. We will also identify stocks that matches the criteria for our “bear” basket and enter those “bear” stocks into short positions. A stock will match our “bear” criteria if a trustworthy source released a negatively-toned story with high probability of impacting its price. All positions are closed on the last day of each month at 3:45 PM.

ADDING WEIGHTS: We add a twist to this strategy by adding weights to some metrics. For example, the more positive the sentiment and higher the impact of an article that mentions a company, the more shares of that company we will buy. The more negative the sentiment and higher the impact of an article that mentions a company, the more shares of that company we will short.

Conditions Used:

Buy if: Article Sentiment > 0.40 and Event Impact Score Entity > 80 and Overall Source Rank > 6
Sell if: Buy if: Article Sentiment < -0.40 and Event Impact Score Entity > 80 and Overall Source Rank > 6

Filtering gif

Clone Algorithm
64
Loading...
Backtest from to with initial capital
Total Returns
--
Alpha
--
Beta
--
Sharpe
--
Sortino
--
Max Drawdown
--
Benchmark Returns
--
Volatility
--
Returns 1 Month 3 Month 6 Month 12 Month
Alpha 1 Month 3 Month 6 Month 12 Month
Beta 1 Month 3 Month 6 Month 12 Month
Sharpe 1 Month 3 Month 6 Month 12 Month
Sortino 1 Month 3 Month 6 Month 12 Month
Volatility 1 Month 3 Month 6 Month 12 Month
Max Drawdown 1 Month 3 Month 6 Month 12 Month
# Backtest ID: 55a6cba7ebe6e20c62dc472a
There was a runtime error.

Hi Kumesh, thanks a lot for providing us with these kind of strategies.
My question: Can you provide us with updated backtests? It would be great to see how these excellent strategies performed since 6/22/2015.

Hi Kumesh,

You seem to be focused on relatively high-beta strategies, which as I understand, will be rejected as candidates for the Q hedge fund. Can they be re-formulated to be market-neutral and still maintain alpha?

Grant

I'm trying the weekly algo but run all the time into:

TimeoutException: Fetcher data not processed in 360 seconds
There was a runtime error on line 112.

Is this algo possible to run live? Alisa?

I haven't tried it :) The algo has 5 minutes to read the CSV before it times out, and it sounds like you're running into this issue. I'd suggest to reduce the amount of data and try running the algo again.

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.

Kumesh, very intriguing results.

I understand this is a weekly strategy, but in my opinion the most effective way to use sentiment to maximize returns is in some sort of daily strategy. Especially with sentiment, the longer you hold, the less returns you will see. Most of our news are published in the early morning of each day and the market will quickly adjust and reflect whatever news information we are making these trades off of. It seems reasonable to assume that we will have maximum predictive power on the current day's returns and that there is no added benefit for the extra holding days.

As to exactly the best way to implement this into a model I'm not sure - maybe 1) a moving average starting when you place the trade and triggering after a decrease of a certain std. dev? 2) a set time to pull out like the original model except instead of pulling out at the end of the week we pull out say 3 hours after the trade or at 15 mins before daily market close? Any thoughts on this would be greatly appreciated!

Also, this is an interesting article for a sentiment-based market-neutral trading strategy which gives consistently favorable returns with low volatility:
http://www3.cs.stonybrook.edu/~skiena/lydia/blogtrading.pdf

Kumesh,

I am a manger interested in alpha stream. Have sent a request for trial through your site but have not received a response.