Journey from investment bank to independent automated trader

[lolatency]

Quote from asiaprop:

absolutely in the next year, no second doubt about it!!! Unless you have access to order flow maybe you can point out where you take your prediction from what the next tick is gonna be and especially how you plan on capitalizing on such knowledge.

By the way, I am not in the business of predicting, I trade AFTER catalysts hit the market, I am no crystal ball reader.

Next year? That makes no sense mathematically. Suppose you have a random walk with a fixed step, that goes up or down one each step. You have a much greater chance of knowing the next step than whatever the random walk will be in one year's worth of steps. Just on account of variance.

Given what you said, if you're in equity derivatives and you know where that price is going to land, you should be making millions already. Just collect the premiums on all the options that you know aren't going to be in the money then.

 

[CloroxCowboy]

Quote from Transatlantic99:

the difference is, at the 50ms end of the spectrum you will be able to more quickly generate a statistically signficant number of trades while being able to recalibrate and refine your model and adapt to changing market conditions...

on the other end of the spectrum, not so much so...

theoretical profits will also always be greater at the high frequency end of the spectrum due to the length of the coastline of the price path, but that is a different argument...

good luck

I will agree that higher frequency systems have the potential to be more profitable than lower frequency systems, due to the longer coastline. I wouldn't say they will always be more profitable, but I trade intraday myself so I agree with your basic premise.

 

[asiaprop]

Partly Incorrect. The variance of a symmetric random walk accumulates at rate one per unit of time and each increment has expected value of zero. So, yes it will be harder to predict the exact location of where the price will be a year hence than where it will end up the next second but I never claimed otherwise. I claim without knowledge of order flow I have better odds of predicting whether the market will be higher or lower one year from now than I have at betting whether the next tick will be an up or down tick.

Quote from lolatency:

Next year? That makes no sense mathematically. Suppose you have a random walk with a fixed step, that goes up or down one each step. You have a much greater chance of knowing the next step than whatever the random walk will be in one year's worth of steps. Just on account of variance.

Given what you said, if you're in equity derivatives and you know where that price is going to land, you should be making millions already. Just collect the premiums on all the options that you know aren't going to be in the money then.

 

[lolatency]

Quote from asiaprop:

Partly Incorrect. The variance of a symmetric random walk accumulates at rate one per unit of time and each increment has expected value of zero. So, yes it will be harder to predict the exact location of where the price will be a year hence than where it will end up the next second but I never claimed otherwise. I claim without knowledge of order flow I have better odds of predicting whether the market will be higher or lower one year from now than I have at betting whether the next tick will be an up or down tick.

Right, so the variance at t+n, n large is much greater than t+1. The price guessing game at very high frequencies is discrete, and not continuous. So you have a very frequent outcomes in a liquid book, really, 0, +1, and -1, +2, -2. +3 and -3 are rather rare, in a book with an average spread of a penny.

Unless you know the up and down probabilities on each outcome, ... I'm not following how you know where you land 1 year out.

 

[asiaprop]

Again I did not claim knowledge of whre I land in a year. But maybe you can tell me how much better you can do than a 50/50 guess on whether your next coin toss will end up heads or tails.

As menioned previously I do not run nor claim knowledge in high frequency trading but I fail to see how you want to come out ahead without access to order flow and assuming you are on the commission paying side.

Quote from lolatency:

Right, so the variance at t+n, n large is much greater than t+1. The price guessing game at very high frequencies is discrete, and not continuous. So you have a very frequent outcomes in a liquid book, really, 0, +1, and -1, +2, -2. +3 and -3 are rather rare, in a book with an average spread of a penny.

Unless you know the up and down probabilities on each outcome, ... I'm not following how you know where you land 1 year out.

 

[academic]

Quote from lolatency:

Unless you know the up and down probabilities on each outcome, ... I'm not following how you know where you land 1 year out.

He claimed "better odds of predicting whether the market will be higher or lower one year from now", not where we land 1 year out.

This is turning into a play on words.

 

[lolatency]

Quote from asiaprop:

Again I did not claim knowledge of whre I land in a year. But maybe you can tell me how much better you can do than a 50/50 guess on whether your next coin toss will end up heads or tails.

As menioned previously I do not run nor claim knowledge in high frequency trading but I fail to see how you want to come out ahead without access to order flow and assuming you are on the commission paying side.

The [conditional] odds are much better than 50/50 on ticks.

 

[asiaprop]

conditional on what exacty? Serious question because I dont have the exposure to high frequency strats that you enjoy at work. I am interested in learning what your edge will be over other guys in the market place who throw tons of bucks at high tech gear, enjoy crisper speeds of execution and lower transaction costs. Can you please elaborate a bit...

Quote from lolatency:

The [conditional] odds are much better than 50/50 on ticks.

 

[lolatency]

Quote from asiaprop:

conditional on what exacty? Serious question because I dont have the exposure to high frequency strats that you enjoy at work. I am interested in learning what your edge will be over other guys in the market place who throw tons of bucks at high tech gear, enjoy crisper speeds of execution and lower transaction costs. Can you please elaborate a bit...

Consider an ARCH(1,1) model. The conditional heteroscedascity has more utility than the unconditional heteroscedascity. The same sorts of arguments apply with tick data.

I can't tell you what I know to minute detail; however, the book "Analysis of Financial Time Series" by Ruey Tsay has a section on high frequency data analysis that sheds some light on what's going on. It won't make you millions (and some of it is not practical), but it has a basic introduction to the topic. Lo's papers have some "Lo" utility, but they're interesting reads.

There's some truth to the idea that papers in academia are junk, but sometimes the empirical studies are useful. Tsay's book is a bit dated, but has some nice stuff to make you get more of a real feel for the market.

 

[lolatency]

Quote from asiaprop:

conditional on what exacty? Serious question because I dont have the exposure to high frequency strats that you enjoy at work. I am interested in learning what your edge will be over other guys in the market place who throw tons of bucks at high tech gear, enjoy crisper speeds of execution and lower transaction costs. Can you please elaborate a bit...

With regards to what my edge is -- the high speed people don't necessarily take certain kinds of risks. I'm willing to take some of those risks that they aren't, and for which their degree of speed doesn't matter. I just have to be fast enough, which is part of my experiment here, not the fastest.