Back to Community
Defining day types to lead strategy selection (Mean reversion vs Momentum)

Hi guys,

This is my first foray into systematic trading so please bear with me if my questions seem basic! I am a discrectionary day trader active mainly in the Bund, Gold, ES-mini and Stoxx futures markets, and have 2 main methods of trading (a mean reversion & a momentum strategy).

Which I implement on any given day depends, at the moment - on my personal analysis which leans heavily on volume / market profile, and an interpretation of wider chart patterns.

Im sure any systematic / quant trader will balk at this, and I have begun to see the faults in my 'lizard brain' when it comes to defining the day type. I know my methods, when used in the right context are profitable as I have been making a living using them solely for the last 18 months, however I know there is room for improvement when it comes to Identifying 'mean reversion' or 'momentum' days, and I feel I need to remove the subjectivity from this area of my analysis.

I am thus wanting to try and work on something that will help guide my judgement on what sort of day is developing (mean reversion vs momentum), whilst also track intra-day changes in this so I am able to react quickly to changes in the environment. At present I am not wanting to automate my trading, but im looking to automate this area of my analysis which I beleive can only add to my effectiveness as a trader!

I am not asking for 'the answer' - I am more than willing to put in the hours , but if anyone could point me in the direction of which metrics I should be using to track and define where a momemtum strategy should be implemented (and vice versa) I would be hugely grateful!

2 responses

First off welcome to Q .

You pose an interesting question which in general is "are there a set of factors which can predict todays 'market type'?"

Maybe use some of the machine learning tools to find out?

I suppose one could define many different 'market types'. However, in order to do any type of analysis, one would need to define these very explicitly. For instance 'momentum day' might be when a group of securities has a high correlation with SPY and SPY return is greater than x. Maybe get more granular and define 'up momentum' and 'down momentum' days?

Once one defines type (or classifier in machine learning terms) the next step would be to brainstorm a number of factors (or metrics in the original post) that one would feel may impact the type. Now figure out how to get all this data into a table. The rows of the table would be trading days. The first column is the type of trading day it was (ie the classifier). The rest of the columns are the data for each of the factors. This last step may not be trivial depending upon the factors. Then, finally, train a machine learning classifier and see which factors predict the classifier and how well the overall prediction is. Q has a good lecture series on machine learning which may be a good start https://www.quantopian.com/posts/machine-learning-on-quantopian

As far as factors to consider, here are some that come to mind. Maybe use ETFs as proxies to develop factors. For example, 'market jump' is simply 'change of price of SPY'.:

  • big overnight jump in the market (open is much different than previous close)
  • big change in first hour trading price and/or volume
  • big change in previous last hour trading price and/or volume
  • direction of bonds vs stocks
  • relative return of bonds vs stocks
  • correlation of sector movement to market (everything moving or just certain sectors)
  • us market correlation with international markets
  • value of dollar vs other currencies
  • volatility change (VXX)

It may be good to test these factors as both binary (market is up or market is down) as well as real number factors (market is up .8%). Don't be afraid to try seemingly unrelated factors. The nice thing with some of the machine learning tools is they can highlight which factors correlate with the classifier and which are simply noise.

All this of course is predicated on the assumption that if one an predict the 'type' of trading day then one can benefit somehow. Before spending time on predicting it may be worth it to validate that assumption. Just a thought.

Best of luck.

Hi Dan,

Many thanks for the response! That makes alot of sense trying to define the day types using some specific factors. I have begun looking at the range of the initial balance, and the probable outcome of the day. Naturaly, a smaller IB often results in more of a trend day as participants are yet to position - however I am still trying to dig into the actual statistics surrounding this.

I understand your concern regarding validating my assumption. Maybe im looking at the problem through the wrong lens - your comments have been very helpful and have helped to direct me towards the actual questions I should be asking!

Best,

J