Smarter, smart-order-routing

One of the algos that I frequently like to re-visit is the one dealing with Smart Order Routing (SOR). SOR is supposed to be one of the simpler, more straightforward trade execution algorithms, or is it? Reality is, SOR algo logic could be more elusive than conventionally thought. The key problem is that "smart" means different things to different participants. Chances are, if you use a sell-side SOR Algo for unmarketable orders, the algorithm will most likely be skewed towards an avenue favored by the algo provider for reasons that might not be readily obvious.

I spent some time evaluating SOR algos offered by sell-side firms for the routing of listed-options orders. I chose options orders because they are slow enough to allow for easier tracking and dispersed enough to generate meaningful realtime observations.

In the U.S. today, active option classes trade on six regional exchanges (the AMEX, BOX, CBOE, ISE, PHLX and the PSE) with the ISE, PSE and CBOE being the heavy weights in active names. I evaluated three different broker sponsored SOR algos for oversized limit-orders and found that for the most part they exhibited similar behaviors and wherever they diverged there was some explanation related to their provider's involvement with a particular execution avenue. The basic logic considers the prevailing spread and size then divvies up amounts accordingly. In some cases, that divvying up takes on an air of rationalization or as I would like to call it "tilting," towards one liquidity source over another.

Everything else being the same where multiple avenues exist you will always come across cases where "time priority" will not be observed for a smartly routed order!

Actually I think tilting is good only if one gets a say into how things get tilted. Which brings us back to why the buy-side needs full control over the Algos they use. Typically, for unmarketable orders, I would throw in extra weighing points towards avenues that show more aggressive market making traits. For instance, the price improvement feature (PIP) offered by BOX would get a heavier weighting (from me) over other avenues with deeper books. I would also tweak the weighing to favor tighter spreads, better bid/offer size ratios and relative participation. In essence I would not treat my SOR as a black-box rather an "adaptive" system, albeit the adaptation in this case is quite discretionary and not rooted in some artificial intelligence logic.

SOR algos are a "must have" for any OMS worth considering, furthermore SOR algo logic must be fully disclosed and allows for user-definable parameters with the ability to dynamically tilt the relative weight of the routing rules based on acquired observations.

In this paradigm, the trader gets to express his/her expertise into re-usable algorithmic rules and that's exactly where we want to be in this brave new world of algorithmic trading..