The Critical Masses

Discussion in 'Journals' started by Ultracruncher, May 28, 2011.

    (Using Web 2.0 to Predict Financial Markets)

    I was recently reading about Derwent Capital. When I say reading I mean mostly skimming headlines and ‘’Key Wording’’articles.This constitutes reading in our brave new world.

    But back to Derwent or the ‘’Twitter Hedge Fund’’ as they have become known as. By taking a 10% sample of daily tweets Derwent attempt to ascertain the mood and sentiment of the population. They believe that by knowing if the cats going to get kicked when you get home may help them predict which way the market is going to go in the next few days. Brilliant!

    After a quick internet search it turns out that quite a bit of research has gone into market psychology, opinion mining, efficient markets and irrational humans. I tried reading a few of the papers but was bored instantly. I expect they discovered that it does not work. That, or they are all billionaires now...But this Derwent thing got me thinking and even kept me up at night for about 5 minutes .

    My big idea is this: What if you could build a chart that uses sentiment instead of price? Yes I know these charts exist already in the form of business and consumer confidence surveys etc... BUT what if you could search for key words across Twitter and Facebook, use Google search volumes and Google Trends and by using different weightings on your data input the values into a model that would give sentiment in a SPECIFIC security a value? Would my ‘sentiment chart’ mirror the price chart? You would think it should. Good sentiment=market up, bad sentiment=market down.

    So after a few minutes of intense work and concentration I developed my trading algorithm which I have rather imaginatively called ‘Fred'.

    What I have found is that the charts I create for a particular security look nothing like their price chart. For the most part they flat line with spikes in either direction. More importantly they spike BEFORE a price move making them a useful leading indicator. Spike up is a buy signal, spike down is a sell signal.

    Back testing is too difficult .I have been forward testing and the results have been surprisingly good. Having the attention span a little less than a kid suffering from Ritalin withdrawal symptoms, I have not been diligent in recording these results. So let this blog be my Official Record. I shall endeavour to post all the trades Fred triggers. Obviously I’m not going to put real money into this because I’m not a complete crackpot! But do join me in this experiment into market psychology and the wisdom of crowds.

    Regards to all,


    (If you are from the Nobel Prize bunch pm me)
  2. Before I post the signals generated by ‘FRED’ I want to go over a few points about the markets that I will be focusing on. Theoretically any and every market should apply. However with limited time resources to cover every market I will be focused on the Dow and Nasdaq 100 constituents.

    Equity signals are generated on a Sunday and for the purposes of this experiment will be hypothetically entered in to at the Monday Open. The position will be closed at the close of the day on which the closing signal is generated.

    The time horizons for the trades are unknown and may last from 1 day to many weeks.

    I should point out that I am not a trader and no discretion shall be used on my part. All entries and exits shall be picked by ‘FRED’.


  3. All the below should be bought on the 31/05 at the open:


    So on his debut Fred decides to generate 7 longs for the same week...I really can’t say I agree as going long equities at this point really goes against my grain.
  4. Sorry, I didn't quite understand - how are you generating the index, exactly? If it's a State Secret, that's cool, too.
  5. web crawler-> selected websites-> data process -> load algorithm-> print chart-> enter trade/exit trade. :)
  6. web crawler-> selected websites-> data process-> load algorithm-> print chart-> enter/exit trade.:)
  7. Longs are as follows:

    AKAM 33.91
    CA 23.05
    CELG 60.25
    JPM 42.91
    SNDK 46.56
    DD 52.69


  8. Missed one...

    CMCSA 24.98
  9. Interesting... Do you plan to publish performance of your calls too?
  10. Hi Sealion33,

    Yes I intend to in percentage form - taken from the Open on the first day the trade is taken to the close of the day the trade is exited.

    As I do not have the time to monitor the markets very closely this is the best and easiest way for me to keep track.Obviously its not the most efficient as trades should be closed immediately after a signal is generated. However my day job makes this impossible.

    I must admit that I could not have hoped for a better Day 1...all 7 well up with CA and SNDK flying


    #10     May 31, 2011