Quote from TSGannGalt:
I guess I haven't posted any tradelog (I actually did post a few but got lost somewhere with all the profanity that came along with it) in the past journal/thread so I went ahead and outputted a tab delimited text file (it's got over 50,000 trades and adding stuff and porting it to xls goes over the upload limit...).
Because Bwolinsky has some odd interest in "Institutional Level" trading... I thought that I should just pull out one of my models I use on BATS (It's on the blotter). It is an intraday Cash Arb model (with Rebate), does around 60-90 trades a day across different symbols. It's not a Single Leg Gamma but it's a pure text book Pairs / Stat. Arb..
This is a live result that's been risk adjusted with commissions. (Posting results with sizing does skew the model). Considering that bwolinsky's model in the first post was about 3 years old, I chose the model with the highest 3 year Sharpe, for the sake of bragging and showing off.
Anyways... you can just drag the txt into your xls and do whatever tests you like to run with it...
Then you can start telling me all the negative things about the result, telling me that it doesn't prove jack shit... (and I know... but I think it'll be fun...) etc. etc. etc.
Quote from Mike805:
I'd also ask how one can extrapolate into the future, from a *limited* sample size, that one is or is not actually dealing with a Cauchy distribution?
Tests for normalcy are qualitative... and, assuming one can convince oneself that the variance is in fact finite, well, even in such a case, how do you determine worst case risk?
Lots and lots of data helps... as does a rock-solid fundamental basis for the model...
Quote from talontrading:
Congrats on your anniversary... have a good celebration.
CFA level 1 is a good start... but you can walk out the door in this city and one of the first 10 people you see will have their full CFA. Half the resumes I get are CFA candidates and I would say 20% are full CFA's. So... I'm not belittling your accomplishment... but these are table stakes and, frankly, not a grand achievement.
Again, not meaning to be insulting, but that's just the truth. And... btw... I'm more likely to hire someone without a CFA than with. Why might that be? (That's a real question and the answer is important.)
Quote from talontrading:
The attached Excel file has 4 worksheets:
Pairs is a daily pairs model that holds from 2 - 20 days. These are returns for the whole system which trades several different stock and ETF pairs.
System1 is a non-discretionary system that I trade.
System2 is a system one of our trainees developed.
Quote from talontrading:
The attached Excel file has 4 worksheets:
Pairs is a daily pairs model that holds from 2 - 20 days. These are returns for the whole system which trades several different stock and ETF pairs.
System1 is a non-discretionary system that I trade.
System2 is a system one of our trainees developed.
Quote from bwolinsky:
ex post edit: On the second pass I realized it's most likely an options strategy, requiring me to re-assess what I think.
I will return to this day after tomorrow. It's a little late, and I was up early, so I'll post back on Friday.
Good Trading,
Good Luck,
Beau Wolinsky [/B]
Quote from bwolinsky:
I think it goes all the way to the explanation you choose to ignore.
As I've said more times than I feel I need to, but maybe I'm phrasing it differently for others to understand:
<b>What you're getting at is covariance stationarity. My pairs model specifically <i>transforms the data</i> into a covariance stationary dataset. Is that clear enough? It can be done in Wealth Lab 4, and Wealth Lab Pro, but the secret requires a few more steps beyond thinking of all the data as a z-score.</b>
If you don't believe you can transform non-covariance staionary data into covariance stationary data, you need a lot more training such as Level II of the CFA exam to test and learn from. I was just fortunate in recognizing at Level I what a "standardized measure of mean reversion" means, and spent the next two years trading after that for the first QID QLD system.
Tests for normalcy are quantitative, but I guess can require some finesse with the interpretation. I think it has helped both of you to see the distribution as what I was seeing. I see a distribution with kurtosis below 1, which I still believe is the actual value excel places on a normal distribution, and the positive skewness. Qualitatively, just looking at the graphical chart of the profit distribution appears as I have described.
Quote from Mike805:
Beau,
One can transform anything into a linear model. This is what turns me off to most (if not all of) published econometric models and why I consider it pseudo-science. Non-linearity is a trait inherent in 95% of all natural phenomina - especially in markets. IMO, simply because professors can easily teach a nice linear model, then create nice exam problems for it, is the reason these models are so prevelant in most cirriculums and also the reason why we see so much literature about them. They're easy to underatand and easy to implement. The real world doesn't work that way, most everything is non-linear...
I'm suprised you dismiss cauchy sets so easily. I think it would be worth everyones time to understand cauchy sets. In fact, they work very well when applied to volatility modeling - easily beating most autoregressive models, ARCH etc. I have done the work and cauchy sets are by far the most representative of non-linear market dynamic, specifically when modeling volatility.
Also, I would think one would need a lot more than three years of system data to determine if the data set you have is normal - quantitatively OR qualitatively.
Mike