A Brief Account of my Background in the Securities Industry
Starting in 1967 as night operator of United Press’ RCA 301 computers where my main duty was to run the accounting programs once a month, I taught myself programming. When asked to join the programming staff six months later, I requested a brief delay to finish the Cobol Flowchart program I was writing in anticipation of a chess program. Instead I wrote the 1968 Election Results Program.
Then I was asked to learn the RCA Spectra 70 and design the stock reporting system for UPI. Some months later, RCA said they hadn’t sold us enough computers to do the job, but I had already convinced myself they would suffice if I used chained Channel Command Words to control the disk drive. Basically, I queued all ticker trades sorted by disk location. Then, for each track, made a CCW chain that read all the stock records, TICing (sending an interrupt) for each so update processing could occur timely, then in the same CCW chain rewrote the stock records and write-checked them. So, for three rotations of the disk, Icould update all symbols on that track that had traded since the last visit to the track. A similar technique was used to dump the appropriate program overlay areas to disk for debugging. (That required gaining control of the operating system, something RCA declined to support, but later asked if they could use.)
The output side of this system was a multi-channel series of newspaper lists of the stock prices. There were 12, 6, 2 and 1 channel services at speeds varying from 66.7 baud to 2400, in Linotype code for feeding directly to the presses. These went out on a schedule, and then, of course as soon after close of market as we received all the trades for the day.
I supervised 14 programmers at the peak of the development cycle for this project, including the 1972 election results system.
After leaving United Press, I developed the stock and commodity data system for the Reuters Ticker, on PDP 8s. During and after that, I created a language to handle the error and correction processing of the still hand-punched commodity market ticker tapes. This approach was very successful and I subsequently sold an implementation on PDP 11/45s to Knight-Ridder in Kansas City.
In 1985 a joint venture with Security Pacific created a bond and text system delivering via satellites using ATT 3b2s on a Unix operating system. And then a multi-terminal bond price contributor system for Market Data, using C++ and for the first time being able to formalize the object-oriented coding methods I had been evolving in C over the years.
And again for Security Pacific, a multi-terminal price contributor system to calculate currency cross rates. This system used all 7 terminals around the world to update the target Telerate page fastest by having each update the next row it could lock on the page.
I’ve lately been developing a genetic algorithm system to search technical techniques for trading system variants for my own trading. I’ve developed a brokerage system to keep track of the traders’ actions, both on Python-based Zope. These are running on two linux boxes for better total throughput.