Whats Stopping Retail Dealers or Banks From Brute Forcing a Clients Profitable Stgy?

Discussion in 'Forex' started by achilles28, Oct 5, 2005.

  1. Okay, since I used to be on the other side (investment banks), something like this was done in secrecy. This is not paranoia, but rather done quite frequently.

    However, it doesn't matter how profitable a *retail* customer is, the amount of gain doesn't warrant a full-blown analysis, a couple of million is not worth the trouble, since a good analysis would cost in the neighborhood of $1M (time, resources, etc). Now, a hedge fund that dominates a product (say Greece government debt) would warrant an analysis.

    A full-blown analysis (flow analysis) is quite time intensive and would often involve 2-3 good quants (phd level) and use of a good compute farm. An investment bank (one that shall remain nameless, obviously) has about 4x32 way servers, and a set of custom written application that can compute the correlation of the trades to about 3,000 variables going back tick by tick over 2 years, every conceivable time period, combination (up to 3-tuple, etc). The quants will then submit the batch jobs to the compute server, which would take (depend on scheduling) around 24-48 hours to run.

    Yes, purely "front-run" the customer trades is strictly illegal (however, it depends on the legal domain, many foreign markets do not have such restrictions), but "flow-analysis" and then would trade when the "customer flow" is predicted to be heavy, technically is not. I was in my compliance officer's office when him, along with the firm's attorneys (one of whom was a former SEC commissioner), deemed this practice "compliant".

    "reverse engineering" happens a lot, for instance, in a CDO product, when a particular bank customer (say) would consistantly buy the high tranches would warrant an analysis as well, trying to figure out why the customer would take on the risk, or do they know a pricing model that the seller doesn't see, etc. The results are typically then taken by the agency and arb desks for the particular product, and incorporate into their flow.

    Being a former analyst (quant/tech), I was involved in this type of analysis four times, twice the results were highly correlated, once was a weak correlation, and once (emerging market) was still a mystery.

     
    #11     Oct 5, 2005
  2. ER9

    ER9

    maybe that would be a good thing. think of all the cash on your side going into the market when your setup appears.

    that might be enough to push momentum and price to higher or lower levels even faster.

    by the time it was widelly discovered and possibly ruined you should have so much cash in the bank it wouldn't even matter :)
     
    #12     Oct 5, 2005
  3. Small trader, retail, even a small manager (< $100M), not worth the trouble. But if the customer is a fairly large hedge fund, fairly large dealer in a particular product, or extreme unusualness, then yes.

    Having involved in four such situations, the categorization of the "target" is very different. One case was a foreign entity (non-government), that was buying up default swaps like crazy (I remember the number was around $10B a week). One case was a hedge fund customer that would at times literally suck the liquidity of a product dry (the fund at one point hold some 40%+ of all outstanding). One case was a fairly quiet customer all of sudden starting to taking on huge positions (even the salesperson was mystified), etc. Only one case that I know of was a fund that did extra-ordinary returns.

    Most of the analysis is done for the bank's protection, liquidity risk, financing risk, counter party risk (in the case of the default swaps, failure to delivery would be hugely problematic). Of course, the results are then fed to the agency and arb desks that can push up / down the pricing models.

     
    #13     Oct 5, 2005
  4. ER9

    ER9

    but if your a small trader taking 1-5 mill a year out of the market do you think anyone would care or notice such a small amount? even if the % return on the invested amount was significant?
     
    #14     Oct 5, 2005
  5. I guess it would depend on the size of the firm. For a ibank, with $50B firm balance sheet, no a few million is not worth the trouble. But for a smaller dealer with, say $500M firm balance, then a few million would be worth the trouble.

    However, a $500M sized firm will also have trouble having quants that can do this type of exhaustive analysis, let alone the technology platform that can do so. The quants will be too expensive for such a firm (a good quant these days cost upwards of $500-800), the compute farms would cost a few million as well. Such analysis are non-trivial, since at the end of the day it would require large computation, whether exhaustive brute-force or extensive MC.
     
    #15     Oct 5, 2005
  6. achilles28

    achilles28

    And you miss mine.

    I’m saying 'what if'. And you're saying 'get real'!!

    I appreciate your selfless attempts to keep me 'grounded', but am not biting.

    I want to step outside the world of a small trader and explore the realm and reality of a big, profitable trader.

    Its a thought experiment. Nothing more.

    You can either play along or chose not too. Impugning me to quit the game isn't going to work.

    The experiment is to imagine arriving at a place of real, lasting, scaleable profitability.

    Assuming that’s the case, and we follow this lucrative system to its logical conclusion - making big $$$ - where does it lead? What kind of environment do we find ourselves trading this system in? Is there a real threat of brokers cracking this system for themselves at this level of trade? etc.
     
    #16     Oct 5, 2005
  7. achilles28

    achilles28

    I know what you mean. But my take on this is 'why fix something that isnt broken'?

    If a person finds an edge, why jepordize it? Short term gains may be higher. But mass front running would likely blow the inefficiency apart.
     
    #17     Oct 5, 2005
  8. achilles28

    achilles28


    Rufus - thanks for taking the time to respond and share your knowledge! Much appreciated.

    I want to respond more later when I have more time.

    Before I do, I want to ask, in your experience, what can a potential target for flow analysis and/or reverse engineering do to minimize, avoid or obfuscate attempts to crack a profitable system - assuming such an analysis has been deemed economically worthwhile for a full workup prior?

    From my understanding of your post, multicorrelational analysis and brute forcing over thousands of variables and time frames makes the statistical concealment of any purely mechanical strategy, almost impossible - from the perspective of a definite non-quant.


    Adding mechanical or adaptive 'false flags' to a system to throw off statistical analysis would likely be uncovered and segregated by multicorrelational analysis into their profitable and unprofitable parts?

    Even 'dumby' signals generated and included in the system at random, could be detected as such and separated - uncovering the genuine strategy bare.

    Countering invasive statistical analysis, one would almost need to hire a professional quant to thwart or minimize the effectiveness of later attempts at cracking the system? Would this approach even work?

    From the other end, restricting inflows of data into the statistical analysis might delay - perhaps not indefinitely - a successful cracking. By this I mean holding multiple trading accounts with different institutions so that open trades are closed out with different institutions - restricting trade information available to the modeling.

    I don’t know. I am completely green to this. What do you think?
     
    #18     Oct 5, 2005
  9. there r ways to do this without necessarily taking any losses, by say placing bids on one ECN and when you get hit, you 'simultaneously' sell / hedge on the other platform, at pretty much the same price level if u pick yr spots, platforms etc well. am not saying its easy tho', just that its doable and it does the job wrt to hiding yr trading pattern.

    additional steps if u really have a hard time sleeping cld be:
    - depending on what sort of vendor front-end system you go for if you trade via API, you may want to consider making a few trades outside of that system, i.e directly via the brokers' GUI
    - if u have a prime broker clearing all yr trades, as u get bigger and profits allowing, enrol the services of a second prime broker, again to split your trades, execs, so there is less to no temptation for any one of them to even try to RE/analyse yr trades (even as you'll end up being medium-term profitable at both institutions). further down the road, there r other 'affordable' set-ups involving setting up a string of BVI private funds or Caymans exempt funds for instance

    this just to say there r ways at all levels to 'always' remain below the radar level if u 'believe' u have reasons to worry about yr arb strategies being front-ran for instance, but rufus 4000 seems to know his sh*t - thanks heaps for the v.expert input btw! - therefore consider the hassle & cost factors carefully and... implement what u feel allows to focus on yr trading best, basically, which i'm sure i don't need me to tell u. cheers,
     
    #19     Oct 5, 2005
  10. Such as the forex systems offered by Advanz? They require that you clear thru gain capital, they know who you are, your account number and which strategies you are running. Even if they aren't playing the other side of the trade, they know where orders are stacking up manipulate the market to some degree.
     
    #20     Oct 5, 2005