Thursday, May 2, 2013

Risk limits are simple and effective

There is so much focus on finding opportunities for a positive E(PnL) (statistically expected Profit & Loss) that risk management research effort has become unsexy. Analyzing historical trading performance, I've noticed that setting daily risk limits would have improved over all returns significantly. The fact remains that effective risk management can actually improve business performance over time.





Positive kurtosis in financial returns

Empirically evidenced, financial return distributions, individual products or spreads, tend to display very positive kurtosis which suggests heavy tails. That is to say, average prices move turn-around points are impossible (for most of us) to predict with accuracy over time, hence the "falling knife" phrase for those trying to buy the bottoms or sell the tops.

Therefore, if a position displays an unrealized net loss and is left alone, the momentum could keep moving until the trader loses his/her shirt. This is how some of biggest losing days I had experienced, and learned from.

Daily risk limits 

Applying a strict daily risk limit is simple to implement, effective over time, and puts you closer to a professional level. That means as soon as open positions reach a certain threshold intraday, everything must be liquidated/hedged to stop the bleeding. Even for buy & hold guys, a risk limit could have significantly reduced draw downs in high volatility periods.  


If this doesn't make sense, then it is pretty obvious that a deep understanding of stats is necessary to become a profitable trader.

Thursday, February 28, 2013

Things I've learned this past year

So it's been about a year since I left the quantitative analyst job to trade full time, here are some things I've learned.








  • It's pretty easy to find inefficiencies for profitable trading, it just takes a ton of work finding the MOST profitable trades. 
  • A good trade must always start with robust theory, therefore when the discovered inefficiency erodes, it would be simple and cheap to stop investing capital into it.
  • Back testing historical data has only 1 real purpose: Estimating transaction costs, explicit and implied, to determine whether the fundamental edge is large enough.
  • About code writing for automation. It more often adds another layer of liabilities than value; therefore unless the trade absolutely requires it, it's better to do everything by hand.
  • Commodity futures/options are the way to go as a self employed trader due to leverage available.
  • Even with profitable trading, life still gives you plenty of other challenges. It never gets easy.
  • There's still a ton of stuff I have to learn.

Sunday, February 3, 2013

S&P500 Daily Return Distribution

Came across a Cook Pine Capital report off my daily reading, and noticed some interesting S&P500 daily return statistics. This not only supports the notion that equity index return distributions are heavily fat tailed, but also that Central Limit Theorem does not always apply in real life.

Description Statistics for S&P Daily Return Data (1927 ~ 2008)
Sample Size: 20,319
Mean: 0.026%
Standard Deviation: 1.182%
Kurtosis: 18.347(!)
Skewness: -0.098

Actual vs. Normal Distributions


Sunday, September 23, 2012

Interesting Trade (purely off historical tests, allegedly)

Came across this trading journal "Universal Trend Trading", where the trader works with Fx and index futures, and claimed
"
...CAGR over last 40 years is about 40%. 
I have a degree in IT, so I wrote my own backtesting engine.
"

Apparently anyone able to write code with minimal understanding of applied statistics could make a killing trading the insanely competitive products. Long story short, his $500K account lost about $200K in the first 2 month, and today (a year later), it is at roughly $190K.


Historical Patterns do not apply for competitive ventures 

Large enterprises invest heavy resources into R&D because they are aware of this. Simply repeating a business model that was successful in the past in no way guarantees similar success into the future.

Again, a good curve fit means nothing

As mentioned by Ramsey Theory, given any large set of random data, patterns can be found. Therefore, it is very unlikely for a purely quantitative approach to succeed in financial trading.


Sunday, September 9, 2012

Nonfarm Payroll Forecast

I keep an eye on economic events often at Trading Economics, and Nonfarm Payroll's one of the more important stats for me. So the next logical though is to apply our own forecast instead of the values from 3rd party websites.

I got data from the US Department of Labor, (link to Nonfarm Payroll data). It is for the period Jan. 1990 to July 2012.


Forecast Method

A simple forecast was done by regression analysis. We match each month's number with the follow, and curve fit the scatter plot where the future value is the Y-axis, i.e. the Independent Variable.Turns out the 2nd degree Polynomial Fit has the highest R-Squared (at 0.67).