• C++ Programming for Financial Engineering
    Highly recommended by thousands of MFE students. Covers essential C++ topics with applications to financial engineering. Learn more Join!
    Python for Finance with Intro to Data Science
    Gain practical understanding of Python to read, understand, and write professional Python code for your first day on the job. Learn more Join!
    An Intuition-Based Options Primer for FE
    Ideal for entry level positions interviews and graduate studies, specializing in options trading arbitrage and options valuation models. Learn more Join!

Looking for a statistician

Which statistical tests are most often used to test and improve automated trading systems?

  • regression coefficient hypothesis testing

    Votes: 0 0.0%
  • non-parametric t-statistics

    Votes: 0 0.0%
  • statistical bootstrapping tests

    Votes: 1 100.0%
  • monte carlo simulations

    Votes: 1 100.0%

  • Total voters
    1
Joined
11/18/11
Messages
8
Points
13
I'm looking for a statistician to collaborate with on an automated trading system that I've coded. It's a Forex scalping strategy I found on another forum where it's interactively traded by some people with seeming success and yet I can't seem to make it profitable over the long term in my backtests. I believe I am missing something and I would like to take a statistical approach to finding what that something is. My statistics knowledge is not advanced, although I do have some understanding from college courses. My programming is on point though (I've been coding for 7 years just about everyday). The system itself is not quantitative in nature and signals are generated based on technical analysis, but I think it can be improved by applying some of these statistical techniques:

regression coefficient hypothesis testing
non-parametric t-statistics
statistical bootstrapping tests
monte carlo simulations

In my research these are good techniques that can be applied to trading systems. I'm looking for someone that can apply these techniques to the system I wrote (written in C#).
 
FX is about speed and volume... scalping is specially about speed and volume. monte carlo simulations take a lotta time to run due due to the law of large numbers... bootstrapping is bullshit but is quick and dirty.. why are you introducing a hypothesis test... and why are you using t tests.... you haven't made it clear what you want to do with these tools.
 
you haven't made it clear what you want to do with these tools.

Thats part of what I'm trying to figure out. I'm only slightly familiar with these tests but I know they can be used to test whether trading signal have statistical significance for alpha generation for example by comparing the results against a random signal generator.

Another problem I encounter is that I can do certain tweaks to my strategy to optimize certain metrics like Sharpe Ratio but it often comes at the cost of reducing or increasing the number of trades so I don't understand how to objectively decide whether said optimization is actually improving the system or not. Also I'm sure there are other types of tests that can be done in regard to money and risk management.

Ultimately, I'm looking to build multiple strategies and diversify with mean reversion, pattern recognition, scalping, trend-following, momentum breakout, and statistical arbitrage systems. I want to avoid HFT and rocket-science ideas (signal processing, highly computational approaches, fuzzy logic, neural networks, machine learning …), Martingale-type allocation strategies, etc

When building a new strategy I want to take an agile approach. For example, code the entry point only and run an “alpha estimation test” that would show statistics about where the instrument’s price is on average 1 bar after, 2 bars after, ..., N bars after the entry point, using Bar.High and Bar.Low. For example, we code a method to recognize a certain bullish reversal candlestick type (long entry order). The “alpha estimation test” would assess the distributions of the P&L should we exit 1, 2, 3, N bars after we’d go long. This distribution would be compared to the distribution if the entry point had been random and the exit point had been 1, 2, 3 or N bars after that random entry. These distributions P&L would then be plotted against user-defined metrics to see if there is a relationship. For example, a metric would be the entry point candle’s range, plotted for each signal against the P&L extremes after 1, 2, 3, N bars. If we can establish a relationship (visually or via R² for example), we could deduce there’s some alpha to be extracted using this new information (the entry candle’s range), and then we can proceed progressively adding things to the system.

But for the time being, I'm looking for someone to collab with that can look at my backtests and identify statistical tests that can improve this first system and then go from there.
 
Back
Top