Trading Catechism

Ok buit my first OpenCL program. This is on my laptop with an Intel GPU. From such humble beginnings bigger things come. The journey of a thousand miles....

The goal is 100k options per second on an FPGA. First I will do Monte Carlo, but eventually I want to do bi/tri-nomial trees.

Code: 
Platforms (2):
    [0] Intel(R) OpenCL [Selected]
    [1] Experimental OpenCL 2.1 CPU Only Platform
Devices (2):
    [0] Intel(R) HD Graphics 5500 [Selected]
    [1] Intel(R) Core(TM) i3-5005U CPU @ 2.00GHz
Build program options: "-D__DO_FLOAT__ -cl-denorms-are-zero -cl-fast-relaxed-math -cl-single-precision-constant -DNSAMP=262144"
Running Monte Carlo options pricing for 65536 options, with 262144 samples
Size of memory region for each array: 262144 bytes
Using risk free rate = 0.05 and volatility = 0.2
Host time: 68.8747 sec.
Host perf: 951.525 Options per second <-- The FPGA version will be 100x more throughput, 
and the latency will be in the low single digit microseconds.
 
There is strong multicollinearity between the independent markets, making it difficult to find good timing. That is not just true because of the high volatility, which usually leads to high multicollinearity, but seems to be true in general these days.

Correlation has gone to 1, without the crash part necessarily.

[As a side note, from a mathematical perspective, it is interesting why multicollinearity is a problem. Almost everything comes down to invertibility: "In case of perfect multicollinearity the design matrix is singular and therefore cannot be inverted. Under these circumstances, for a general linear model {\displaystyle y=X\beta +\epsilon }
b4b37cd053fae656404e567945563b5f80630073
, the ordinary least-squares estimator {\displaystyle {\hat {\beta }}_{OLS}=(X^{\top }X)^{-1}X^{\top }y}
f067e916bb85c9f0032d081fc22305f414809e3d
does not exist."]
 
Who Made Money in the Brexit Chaos? Machines, Not Humans
In the weeks leading up to the Brexit vote, the trading models at many firms adopted a defensive pose.
By
CAROLYN CUI,

BRADLEY HOPE and

GREGORY ZUCKERMAN
Updated June 29, 2016 10:19 a.m. ET


Wall Street is tallying up the winners and losers after the severe market reaction to last week’s Brexit vote. One theme has emerged early—the computers got it right and the humans got it wrong.

http://www.wsj.com/articles/in-brexit-trading-machine-beats-man-1467158146
 
20160629_151153.jpg
The SoC is a little smaller than my mouse but packs a great learning FPGA. All you need is a bone dry cappuccino, and we are ready to dive in and learn.
 
This little SoC has so far given me nothing but pleasure. The SoC is running a web sever. When you browse to the SoC, you can do all sorts of things instantly that manipulates the board through the FPGA. Simple things like manipulating LEDs, but this gains you confidence.

Part of the system shows up on Windows 10 desktop as a "disk". There you install the drivers by double-clicking on the 32/64 bit version of your host OS. It took five seconds. After that, with the SD card already installed on the SoC from the factory running Linux, I guessed that it was running ssh and VNC. I guessed the VNC address and it immediately came up!

Ordered the AdaFruit TFT display so that I can run this and other demos. Incredible the FPGA speedups, as you can see when the demo switches from FPGA to CPU:

 
Last edited:
You must log in or register to reply here.
Back
Top