Alpha One Intraday Long-Only NASDAQ 100 New/Blog Backtest [No Leverage]

What type of sources are you using for your data feed?

Hi Daniel,

We're using over 20 million news and blog sources across the web. Any sources that are publicly accessible to someone on the web, we're tracking. For example, from your premium news sources such as Bloomberg and Reuters, to your trusted blog sources such as Seeking Alpha or StreetInsider. Also everyday, we automatically detecting and adding new sources to our analysis.

Furthermore, we're able to track and find valuable low-traffic web sources. Some of these low-traffic web sources may have valuable information before any other major new sources but not many individuals know about them. When these low-traffic sources release valuable information, they are not indexed by major search engines in a timely manner whereas when major news sites such as bloomberg release information, they get indexed in seconds and are searchable via search engines.

To solve this issue, we are constantly monitoring these valuable low-traffic sources frequently and once valuable information are release, we analyze it and send the analytics to you.

Hope this helps!

Best,
Kumesh

Thank you for the prompt response, Kumesh. We have spoken before, but I was curious as to how you implore your data analysis to the quantopian-minded entrepreneur. Do you build/utilize algos for your platform specifically or are you an API/Data Feed that is able to plug into any algo?

Ah, no we're just a data feed provider! You can use our data feed to plug into any of your automated trading models. The only reason we create algorithms and backtest reports is to make your lives easier if you decided to try out our feed.

Have you backtested your algos to run from 2002? I would like to know how they did during the 07-09 markets.

Hi Dan, we only have data going back to 2012... it would be interesting to see though, even for myself. We'll try to extend the history in a few months.

Hi Kumesh, Per your original post,

Impact: Identify the probability a company's price may get affected on the same day.

Yet per the store page (and it's respective notebook), it says

impact_score - on [0,100], this is the probability that the stock price will change by more than 1% (given by: close - open / open) on the next trading day

The AlphaOne.csv file you provided (thanks!) has start_date set as a date-only field (no time stamp), while other datasets (such as what you shared on the Twitter Leak post) has start_date with granularity down to the minute. Is the date of applicability dependent on the start_date granularity? (ie if we see minutes, then the data is a stream meant to be used real time....but if there is only a date present, then it represents aggregate data collected on that date and to be used the following trading day).

I ask because if the start_date field in the data attached to this post is meant to be used the next trading day, then I'd expect to see a time shift in the algo when pre_func() creates date_new. Not shifting occurs, which I believe would cause look-ahead bias. If the dates in the CSV data are already shifted, then it looks good!

(Anshul from Accern here) Hi Jason,

Alpha One dataset is aggregated and is delivered daily, as compared to Alpha Stream (our product for institutional clients) which is a real-time quantitative data feed. The twitter leak case study is based on Alpha Stream data, which is why you can see minute/second level granularity. As Alpha One data is aggregated daily, you only see a date column. Another point to add: The "date" column in the Alpha One dataset is already shifted by a day. For example, the Alpha One data you receive today (10/21/2015) on Quantopian will have metrics aggregated in last 24 hours (10/20/2015) but today's date in the date column. This is what we meant by next trading day. We wanted to simplify this so traders don't have to intelligently increment the date in their strategies. As you can see, there is no look-ahead bias. In the original post, Kumesh was referring to the date column itself, that's why the mention of same day.

I agree that it is a bit confusing, especially the mismatch between date description. We will make them consistent. Thanks again for pointing this out.
Please let me know if you have any other question.

hi Kumesh, i know there is accern version 2 data set. For this report, are you using original or version 2?

Sophie,
At the time this post was written, Kumesh used fetcher to import the data.

Now, this data is available built into the platform. See the Accern Alphaone detail page for more information, including links to algorithms that use the new built in data.