thanks for the link. I'm quite familiar with execution speeds etc. I was just curious as to how fast do trading firms' option pricing algos work. In other words, how long does the program take to process new information and spit out a new quote. The bottle-neck is in no way execution since this can be done in fractions of a microsecond as the article clearly points out. The bottleneck is in the option pricing algo which I'd like to beat thanks!
My takeaway from recent conversations with options MMs (I used to work for one of the big MM firms but am not active in the business now) is that Greeks are cheap/fast -- it's just arithmetic, poly approximations multiplied and added -- whereas the problematic bit is communications -- both latency, and variability-of-latency.
Rodney King is exactly correct. From the time we get an underlying update we can reprice an entire option chain in 2ms, and send out new quotes in another 2 ms. If you want to add safety rules, and inventory adjustments it takes another 2 to 3 ms. However, if you are using Opra/Siac data you will be behind a company that pulls directly from an exchange. This is especially true if you are waiting on stock data from NY, and quoting options in Chicago. The OP mentioned doing futures AQ'ing so perhaps you are getting underlying updates from the same exchange to which you are quoting options. This should help immensely. However, there are a slew of other issues involved. For instance, most exchanges limit the number of quote changes you can send in a second, while also requiring minimum width and minimum chain % rules. This forces the MM firm to use their quote changes efficiently, which is not easy. Also, most exchanges will have some kind of quote packet cue which you must deal with. This means that when an underlying change occurs and every company sends new quotes, the exchange will update the quotes in the order they were recieved. If the exchange is currently bombarded with quote packets, it can take them up to a second to change your quotes, despite you sending these quotes to them in 8 ms. Therefore, you will need a ton of safety measures, and must learn to use the various quote canceling mechanisms properly. Here's the real rub: no one will teach this stuff to you. The Api people at the exchange typically don't know as much about their own systems as the trading companies. The trading companies only learn what works by losing money. Therefore, they will not give you their knowledge that cost them millions to learn. If you are a service providing company that doesn't trade you will have zero chance of creating an efficient AQ system--at least compared to those companies that have their money on the line every day. Conclusion: building a successful AQ system is one of the most difficult technical problems to tackle. I would say it takes a minimum of 4 senior level guys working for 2 to 3 years, at a cost of $5 million (including trading fees, losses, etc.) to build successfully. Most companies (even large investment banks) that begin this process will vastly underestimate the costs and difficulty, and in the end don't successfully complete an AQ system.
That's a great post from Mr. Store. I'd add that the market-cap of an options mm firm doing X% of total options volume is likely much smaller than that of a similar firm doing the same X% ten years ago -- which is an indicator of how tough the business is. Of course, market cap is tough to gauge, since the firms are privately held, but that's a reasonable guess based on public transactions -- acquistions & etc.
After going over the new SEC Market Access requirement, I believe that complying with the new regulation will *significantly* increase order latency. According to this http://complianceinsights.typepad.c...ec-rule-15c3-5-supervising-market-access.html: "These firms will be required to ... establish, document, and maintain a system of risk management controls and supervisory procedures reasonably designed to manage the financial, regulatory, and other risks of its market access business. The Rule applies to transactions by all B/D's with market access, and does not distinguish between transactions for a B/D's own prop account - including market making activities - and traditional agency transactions." I don't see how this regulation doesn't add huge "pre-trade" processing requirements and corresponding latency for option market makers to firms that properly adhere to it. If you read the regulation carefully, it implies a completely independent layer between the quote server and the exchange that can evaluate and intercept any order that is deemed to violate the established risk parameters. Is 10ms latency even possible in this sort of environment? I know that the implementation has now been backed up to Nov 2011. However, if market making firms take this regulation seriously, I think we are going to see a huge impact on market liquidity as the marginal market makers are forced off the air. Comments?
Now you want to AQ in options? When they've gone from a quarter to a penny? Right now the atm options on PCLN are tighter than the shares. I suggest you look to algo-trading in delta-one and forget the options.
Right, bc the order hits the DPM at the post and then they hedge in spot. I wonder how many firms make their living front-running the spot hedge?
Thanks for replying, hippie, I'm not sure I fully appreciate your point about options occasionally ticking ahead of a stock. Jerkstore, above, gave a ballpark guesstimate of <10 msec for the best architectures to paint an option chain in response to an equity price tick. The point of my post is to question whether this time frame is realistic in an architecture that complies with the pending SEC Market Access regulation.