hypothesis driven vs ad hoc

Discussion in 'Strategy Building' started by ducatista, Mar 18, 2020.

  1. ducatista

    ducatista

    sometimes whilst combing through the data researching a potential strategy, ill find data that directly contradicts my original hypothesis. for example, lets say you hypothesize a popular pattern/whatever will return a positive expectation due to some fundamental explanation, but upon further research you see it actually returns a consistently negative edge.

    eg you expect some long market pattern to be consistently profitable by $1, but you see it actually consistently loses $1. does this mean it could make a consistent short opportunity?

    NOW WHAT IF you previously held some sound economic theory behind why said pattern works, but find out it behaves totally against conventional logic and you cant figure out why. say results are consistent over a ten year period. would you take this trade even though it swims against conventional logic? or if you cant derive some logical reason for this edge exists

    hope ive explained this well enough
     
    .sigma and tommcginnis like this.
  2. No

    Logic plus evidence. Both required. Every time.

    GAT
     
  3. Perilace

    Perilace

    On the one hand, I always say that it is necessary to stick to the system and always act strictly according to your strategy. But I can confess absolutely honestly that sometimes there are really cases when you see that everything goes wrong and you continue to break the rules. I can't say that there is a certain explanation for this, I would say that it is a kind of instinct that just guides you through the market. I think that many people are faced with such issues and events, because from time to time no system works and you have a clear understanding of what to do next. I think that experience probably plays a role when you start to understand something that was not available to you before.
     
    .sigma and tommcginnis like this.
  4. ducatista

    ducatista

    how about ml traders
     
  5. Do you mean machine learning?

    Most good ML quant traders still operate using the idea of having a hypothesis to test before looking at the data. There are some domains where that doesn't make sense, but for most it does.

    GAT
     
  6. tommcginnis

    tommcginnis

    I have had too many instances of unsupported "reactions to data" being profitable, and likewise, too many instances of neoclassically-grounded trading notions/ideas/theories/rule-based_trades/algos behaving without *any* regard for the data from which they were formed.....:confused:

    GAT's observation that trades should always having a grounding in evidence&theory first -- *that* is the preferred path, for sure. But this is not a dichotomous choice, either [":thumbsup:"] : easily a third of my trading agenda is following data, discerning profitable paths, and finding/testing explanatory hypotheses around it. "Not quite a thesis," but a hypothesis.

    A little risk control, a little data, a pint of beer on a sunny porch.....:p A wonderful thing.

    Not exactly as (he) intended, but to borrow from Thoreau, "If you have built castles in the air, your work need not be lost; that is where they should be. Now put the foundations under them."
     
    .sigma likes this.