10
$\begingroup$

have been asked to look at some financial timeseries for potential suspicious activity. These are stocks (my background fixed income hybrids trading and not forensic analyst...) and most of the conclusions will be drawn from granular trade level data (spoofing etc).

However, there exist general quantitative / statistical approaches to detection of patterns of activity and informational asymmetry and am wondering if anyone can give me some references (regulatory or otherwise). I found this general survey but would like pointers on journals / books / papers to consult if possible:

www.consob.it/documenti/quaderni/qdf54en.pdf

(also I think Algorithmic Trading by Cartea et al has some paragraphs on infomational asymmetry, which will explore).

There must be a substantial body of research on this, am hoping some might be public domain, can anyone help?

Thanks

EDIT:

There's some interesting pattern recognition work if interested: https://webdocs.cs.ualberta.ca/~zaiane/postscript/DSAA2014.pdf; http://www.ijtef.org/vol7/503-FR00023.pdf

$\endgroup$
2
  • 1
    $\begingroup$ You shouldn't be thinking about quantitative approaches as the first step. This needs to be hypothesis/market expert driven with the appropriate method following from this. $\endgroup$
    – Jase
    Feb 18, 2017 at 4:16
  • $\begingroup$ @Jase for sure - actually in my case there is a clear hypothesis of a pattern of behaviour in conjunction with various events, the quantitative side is just to lend further struts to the case. $\endgroup$
    – Mehness
    Feb 28, 2017 at 18:24

2 Answers 2

6
$\begingroup$

I can't help as much with public literature, but I did see a talk with a member of the FINRA data science team responsible for exactly this (event link below - perhaps you can track down the speaker).

I don't know the structure of your data, but the approach FINRA took was to develop trader-level attributes (not stock level) to create profiles for each trader (I.e. Number of trades, volume, etc.) and applied unsupervised clustering algorithms to the traders. Once they did so, they determined which neighborhood clusters seemed okay and which were large outliers / potentially malicious. You didn't mention if your data set had trader id's, but this would likely be my approach and FINRA seemed to have success.

I hope this helps and good luck!

http://d1ryye6yw47pmy.cloudfront.net/images/3651/2016_Trading_Event.pdf?1463507984

$\endgroup$
3
  • $\begingroup$ That's extremely helpful thanks very much. Part of the task is to specify exactly what kind of data will be required actually - clearly trader_ID level granularity would be excellent, however failing that I think there will still be inference possibly from anonymised trading patterns. I have been able to find some work (in part done by the author of the paper I've cited above) looking at this kind of problem. Ideally I'd like to be able to reference 'received' wisdom / regulatory financial 'tripwires' in addition to independently conceived analysis to strengthen the case. $\endgroup$
    – Mehness
    Jan 18, 2017 at 16:04
  • $\begingroup$ Thanks again for the FINRA link - will explore. If I find useful public domain stuff will post above, cheers $\endgroup$
    – Mehness
    Jan 18, 2017 at 16:04
  • 1
    $\begingroup$ This is actually the same approach some insurance companies take to fraud. You should definitely use all the information you have though - if you have the addresses of the traders, then convert them to long/latitude and look at distances, if you can link them to individual people and can track moves of people between companies, that'll help too. The fraud approaches seem to be built onto of building a graph of relationships, and finding statistically unlikely structures within the graphs. $\endgroup$
    – will
    Feb 18, 2017 at 14:13
3
$\begingroup$

I cannot speak specifically because I have proprietary insight into the issue, but one approach is to compare the timing of trading activity (volume, bids/offers on the book traded and withdrawn) with publicly-announced news.

Think Granger Causality, and which event seems to cause the other.

$\endgroup$
2
  • $\begingroup$ Thanks for this - do you have any pointers as to where I could get hold of such data? Free is best however if it costs, it costs, thanks. $\endgroup$
    – Mehness
    Jan 24, 2017 at 17:13
  • $\begingroup$ I often hear, "I wonder if there was a news event that caused this move." What you will often find, however, is that some people were already trading on the news before it was broken, and I'm not talking about the hunches retailers might have. However you can get your data, @Mehness, I think you'll need tick data in order to get more accurate bid/ask volume. I use NinjaTrader, and they do have some free data. $\endgroup$ Sep 16, 2021 at 19:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.