Tuesday, 13 March 2012

Using Algos Intelligently

I have a bunch of algos that I run on different markets. There are a number of algos for each market and I use the one that is appropriate to the context.

How do I know in advance which one to use? I can't know for sure but that doesn't matter because each algo I have is historically profitable over time.

Some algos are most profitable in very trending markets, others are better in more choppy days.

I manage my algos just as I manage my trades. I know what announcements are due during the week. I know when we are in earnings season. I know when the Fed will talk. There are lots of things I can know in advance if I do my homework.

I turn on the algo that I think will best suit the vision for the day. I also break down the days into three time zones and may have different algos, or the same algo with different settings, running during the day. If my vision is wrong I can switch algos.

The strength in running algos is that I can be more diversified and do not need to be glued to my screen.

Having the same algo with the same settings running all the time can be profitable but nowhere near as profitable as doing the extra work to mine the data so a more intelligent use can be made of the algo.

Today's DAX during U.S. RTH was trading in the smaller range as has been the case recently. I've reduced the range bars to fit the volatility.


  1. Hi Tom, I think this is a fantastic summary for what you do and have much the same approach - a blend of discretion then execution using algos.
    I would like to ask, your comment, "each algo I have is historically profitable over time" is that just through history showing its profitable (a leap of faith if you like) or via backtesting other than a walk through whereby you switch them on and off. It one of those things I still struggle with about the faith in a system. (I use similar methods except I use Donchian channel mid points rather than moving averages to enter with mini trends). Thanks

    1. Anon 01:22, Firstly, the algo has to backtest as profitable monthly over the last 6 months during the whole major trading session eg DAX 7am to 6pm London time. I build the algo with certain market characteristics in mind and then run it when I anticipate or see those charcteristics. In case I'm wrong in my algo choice, it should still be OK over time although may not be the best algo choice for that type of market.

  2. Thanks, a bit as I suspected - the algo must work (profitably) on its own, its just with a bit of discretion it works a lot better. the Context is key, eg; running a long breakout system when a long term downtrend is in place and you are clearly entering resistance may not be ideal :)
    I am yet to get faith in my systems enough to fully automate them because of this dilemma (too many stops, not getting reentry again etc;) of being unable to program discretion/context - I guess its good to see others doing the same thing. (While I still have the dream of turning on the computer and coming back to a large account :) without me interfering).
    Thanks Tom