Algorithmic Trading

Discussion in 'Strategy Development' started by Scottsdale, Aug 18, 2005.

  1. It seems like more algorithmic market making is taking place on the Nasdaq, NYSE, and futures exchanges than ever before. Seeing as how algorithmic trading is reactive rather than proactive, a sharp human trader should be able to pick up on some variables that the algorithms are triggering off of. I have successfully scalped US stocks since 1998 (first Naz, now NYSE), but I cannot seem to pick up the logic behind today's market making algorithms. Has anyone identified any logic behind the randomness? Let's get the ideas flowing and hopefully find an edge that we can exploit!
     
  2. Good and interesting post...lookks to me like most are doin rebate , reversion to the mean type bidding and offering

    i have also been trying to see patterns in nazdaq stocks...

    u'll notice if there is a nice spread and you low offer (or high bid)

    may OTHER ECNS will join you

    on NYSE you'll notice below NYSE there wil always be ecns bidding and offering that too wil join eachother ONLY TO a certain number...then if you go high bid they stop

    i think lot depends on Avg Trading Range, Avg Trading VOlume, etc

    more ideas would help

    d
     
  3. nitro

    nitro

    Why don't you start by expanding on how you successfully scalped NAZ in the past, and NYSE now?

    nitro
     
  4. Lots of mean regression ones that are continuing to dampen volatilites. There are tons of algo's out there that essentially use the ecn's to play market maker scaling in and out all day long....only unlike the registered market makers who still have to maintain two sided quotes (even thought they can be a mile away from the inside)....these programs trade with the odds stacked even more in their favor technically speaking since they lack the access to order flow information that tradtional MM's can still feed off of.
     
  5. I haven't traded the nasdaq since the pre-supersoes/supermontage days. On the NYSE I read the tape and open book for clues about relative strength or weakness of the stock relative to the sector and overall market. I try to step in front of size early in a move, and fade it when the move begins to slow after overextending itself.
     
  6. cosine

    cosine

    Seeing as how algorithmic trading is reactive rather than proactive, a sharp human trader should be able to pick up on some variables that the algorithms are triggering off of. I have successfully scalped US stocks since 1998 (first Naz, now NYSE), but I cannot seem to pick up the logic behind today's market making algorithms. Has anyone identified any logic behind the randomness? Let's get the ideas flowing and hopefully find an edge that we can exploit!

    I guess it would depend on who wrote the algo, and how in depth they went in order for it to be successful.

    Idiot algos compete to take the best possible bid/ask up to a certain point. They get involved into idiot price wars where you can obviously force them into selling/buying you at a more competitive price then they should. No need to develop how this can lead to losses for the banks who run them.

    Smart algos try to statistically infer the probable value of an asset given a certain trade and its characteristics. This way, they can quote the best possible price and control their expected profits (and avoid expected losses). They also perform inventory optimization so that they will privilege trades that can help them rid of their excess inventory.

    Playing with smart algos requires you to understand how they statistically infer the value of an asset given the trade. Although you cannot profit from their lack of precision just by "noise trading", you could find profitable strategies to trade large volumes based on some information without affecting the quotes too much. Obviously, such strategies are used at their best when performing arbitrages.

    These are my thoughts. Welcome comments and additions.

    cosine
     
  7. Mean reversion may be profitable the vast majority of the time, but it only takes one crazy day to negate many slightly profitable mean reverting days. This is especially true in the nasdaq stocks with small daily ranges, and .01 spreads. Access to order flow information is by far the best edge there is in the market place. I can't imagine why anyone would use the ECN's to make a penny wide market without having access to orderflow information. Furthermore, anyone who does have orderflow info doesn't really need a black box to exploit their edge to the fullest. The black box only really saves trader salary, and allows more stocks to be covered.

    The real questions are: What paramaters are plugged into the algorithms to determine when to scale in or out? How large of a maximum position is taken? How frequently do the parameters change?
     
  8. Idiot price wars is one of the best descriptions of the nasdaq that I've heard! On the other hand, this phenomenon has become such a dominant force in the market, that I sometimes feel like an idiot for not making more of an effort to exploit it.
     
  9. nitro

    nitro

    Ok.

    And what edge do you think you can have over machines that are basically making markets based on their current baskets/inventories and where the ES/NQ/YM are trading? In other words, even if you knew the exact algorithm by which these things were basing their trading decisions on, how would that make you a better trader?

    What I trying to point out is that understanding the machine will only help you in programming another machine maybe try to do the same thing, but I doubt it would help you as a human being at all. The way you trade now is the only reasonable way, outside of technical analysis, that a human could trade.

    There are only two ways to trade outside of arbitrage (statistical or otherwise), to take liquidity away from the maket and therefore pay the B/A spread (for whatever reason, breakout, etc), or to add liquidity by making a spread, which is mostly what computers acting as MMs are doing, except when they come in a buy/sell baskets for index arbitrage where they are then taking liquidity. The MMing is almost 100% the domain of well capitilized MMs.

    People like Bright traders try to do pseudo MM type strategies like the Opening Strategy, or the Trade Through strategy but that is a huge patience game and is not for everyone.

    What I am trying to get at is that, imo, very few of the players can play these games, although three people that I know of on ET claim to be able to trade stocks 100% by computer, and no one in their right mind is going to tell you how they are making money doing that (if they are, which is a big IF)

    Most of the reason the markets seem so confused now is that there are so many different ways that stocks get kicked around by their derivatives, whether it be Futs on them, Futs that include them, Options on them, or ETFs that include them.

    nitro
     
  10. Take your successful scalping to the mext step using your present inquiry orientation.

    Now, you do not have a method to get in and be "pushed". What you are looking for will not give you a way in either.

    So plan on using the stuff that you will understand soon as a result of your search being completed to prolong your scalps into real money makers.

    There is one terrific place to "see" what you are looking for. It was thoroughly explained by AMTSWA (sp?) vis a vis tape reading (DOM/button).

    The reading or mechanical equivalent reading satisfies (answers) two future Q's (of three) that will enable you to understand the three elements (there, missing and telltale)

    1. What is the real portion of what is there?

    2. What does what is missing tell you about 1?

    Since you cannot enter knowledgeably, and since once you are in, you are lagging the market (the source and cause of your inquiry), using what you learn will allow you to differentiate between a scalp (little money made if even possible with high fixed costs) and a knowledgeable exit (relatively high reward/low risk money velocity).

    When you get to the knowlegeable exit, then two things will happen.

    1. You will have been in the market and making money already for some of your prior scalp entries (this eliminates a lot of a prior fixed cost for you).

    2. You will begining to recognize knowledgeable exits as another trading facet for making more money. They are yet another set of replacements for your present entry stuff. Any knowledgeable exit may, at some point in time, be seen as a possible very low risk entry time since there was just the conclusion of taking profits that defines the exit.

    All of the above puts you in the market for greater percentages of total market time. This, time in the market during changing price, is a necessity for making money. Being in the market during the time when other participants, through the use of algorithms, are changing their orientation, is a high probability time for change of price. Durations of change of price are singular periods in contrast to those of periods of end of change of price which involve either two or three events happening in very rapid sequence relative to your personal speed (non singular periods).

    Almost all successful scalping occurs in singular periods and almost no successful scalping occurs in non singular periods. Scalpers do not differentiate between these levels of complexity as you know and is demonstrated by your inquiry and the responses that it yields.

    The search for anomolies centers on non singular events because of the abundance of combinations and permutations of market facets. This turf is renowned for high fix costs because of the short duration of possible profitmaking by the very nature of anomolies. It is a self screwing outcome, in other words.

    Your quiry is a demonstration of how to move away from anomolies and towards being pushed by the actions of anomoly utilizers. Your key actions when you learn to manage this stuff will be to treat the pertibations of anomoly "actors" as an advantage to drive you across their riffles to take advantage of the total current flow of the market bank to bank (within its current containers (read it both ways)).

    There are three mathematical "effects" at play here (from simplest to more complex): gradient, divergence and curl. A former French poster in ET would acknowledge this smartly and thoroughly. The algorithms' propensities work inversly with respect to the importance of the market flow variations.

    Your quest can take you to be able to realize (make money) on the more important opportunities and to not continue to be trapped by the incidental anomolies.

    There are very few threads in ET that look into sizing up how market maths can be used to focus on more or less important things in profit making. To reach the compromise among the money making factors, it is an interesting consideration of fixed costs, anomolies and being on the real right side of the market. Optimization takes these out of the picture in the order listed.
     
    #10     Aug 18, 2005