3
$\begingroup$

For a relatively small subset of stock symbols I have been able to build a model that is able to 20-100 times per day consistently predict whether a stock is going up within the next 2 minutes, being correct 75% of the time. I have worked both academically and professionally with data science, and dare say I am using proper methodologies, i.e. not testing the model on training data, keeping in mind that I'm working with time series, etc. So for argument's sake, let's assume that this performance would translate into production.

But if I was to go trade stocks directly based on this model in real life, e.g. through Interactive Brokers, the commissions and bid-ask-spread would more than eat the profits.

As an example, let's say that the typical spread for "SymbolA" is 10 cents, and ignore commissions for now. My model might be able to tell with 75% accuracy whether the stock will move by more than 5 cents within the next 2 minutes. But the model doesn't maintain its accuracy if it is to predict the same for movements of 10 cents, for instance.

So for now this is a roadblock for monetizing this. Therefore my question is: How can I monetize this ability?

E.g. Is there any way of gearing these movements without gearing the spread too, so to say?

From searching, I've stumbled upon (60-second) Binary Options, but I can't figure out of these also incorporate the underlying asset's spread (and commission), by setting the bar for a "win" relatively high? But on the surface it does seem that they forego spreads and commission, by rather paying out just \$0.7 on a win and -\$1 on a loss. So seemingly I could be profitable using these (E[75% * \$0.7 - 25% * \$1.0] = $0.275)

But maybe I'm misunderstanding how binary options work.

Any input would be appreciated - thanks!

Improve this question
$\endgroup$
  • 5
    $\begingroup$ It is not uncommon to find predictive capabilities via ignoring the facts of the market; often the reason for the exploitable alpha is exactly that it is impossible to exploit for anyone (in this case, predicting values smaller than the spread is pointless for, I think, everyone). $\endgroup$ – Forgottenscience Apr 18 '20 at 15:35
  • $\begingroup$ Hi: Have you tried the same framework-concept on a longer horizon ? As it stands, it can't be monetized as far as I know. The only way to make more than the spread ( in a non-HFT framework ) intraday is to use horizons of atleast 10 minutes or more. Even 10 minutes is often not a long enough horizon. $\endgroup$ – mark leeds Apr 18 '20 at 22:25
  • $\begingroup$ Thanks for both comments so far. @markleeds Yes, I actually built the model to predict for a range of various timespans into the future (1-20 minutes). And what I saw was that while the movements indeed got larger with longer time frames, the accuracy dropped significantly as well as the frequency of those events where the model was predicting a movement was dropping too. So all in all, this trade-off between accuracy and potential win still didn't make it above the costs of spread and commissions $\endgroup$ – Blue Swan Apr 19 '20 at 16:11
  • $\begingroup$ yeah, that sort of makes sense because opportunities in that range would probably go away pretty quickly if they existed. One question if you don't mind: What do you use for commissions and do you assume that you buy at the ask and sell at the bid ? Thanks. $\endgroup$ – mark leeds Apr 20 '20 at 17:41
  • $\begingroup$ @markleeds apologies for the late reply. For commissions I use what Interactive Brokers have listed for where I'm from and the type I intended to trade (interactivebrokers.com/en/index.php?f=1590, either Tiered or Fixed, not "Free") . And for the price I assumed I'd sell/buy at, yes, I would assume that. Perhaps that was even a bit optimistic, because I think there'd often only be 100 stocks available at the bid/ask, no? (am thinking of if one takes a look at an a visualization of an order book, usually there's only very few stocks offered at the mid-price) $\endgroup$ – Blue Swan May 11 '20 at 15:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.