BWolinsky Trading

Status
Not open for further replies.
Funny, I wrote nearly the same post and deleted it after I posted it because I figured if we were stuck on skewness and kurtosis, talking about stationarity and transforms was kind of pointless.

This is the problem, the kid half learned some stuff in a few classes, has a mediocre retail testing platform and is talking himself up to build his confidence because he thinks that's how you make it.

I certainly don't want to dampen his enthusiasm... I'm trying to remain optimistic about the outcome of this discussion so hopefully we can get something constructive together.

I'd be happy if I could just get him to stop using Excel to analyze data... Baby steps... :)

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
More...
 
Quote from TSGannGalt:

Good posts. I've overlooked Cauchy, myself. Got some new research to do...
More...

The deal is I do have a life outside of here. It is not completely on the internet. Tomorrow I will complete my analysis. There were other topics of conversation of far greater importance on collective2.com this week, and I was mainly posting there.

Sorry for the slow turn around. It's coming, and, Gann, I've already told you I can't analyze your data without it being placed into a spreadsheet. If Talon doesn't want me to use Excel, I'll use STATA 9 Intercooled.
 
Quote from talontrading:

Funny, I wrote nearly the same post and deleted it after I posted it because I figured if we were stuck on skewness and kurtosis, talking about stationarity and transforms was kind of pointless.

This is the problem, the kid half learned some stuff in a few classes, has a mediocre retail testing platform and is talking himself up to build his confidence because he thinks that's how you make it.

I certainly don't want to dampen his enthusiasm... I'm trying to remain optimistic about the outcome of this discussion so hopefully we can get something constructive together.

I'd be happy if I could just get him to stop using Excel to analyze data... Baby steps... :)
More...

You vastly overestimate the capabilities of someone that has yet to create a profitable trading system. WL is a toy to some, but only to the untrained. WL is also the only platform in existence with a portfolio level simulator advanced enough to not require vast modification to calculate simulations. WL5 .NET is a much more versatile platform. That in itself, is beyond TS, NinjaTrader, and E-signal EFS2 JavaScript. I'll use excel as well as STATA and we'll compare the two. I don't believe my analysis would be any different.

I don't particularly care if you think I know stats or not. I'll show you a quantitative analysis model I use on occassion to calculate the fair values of stocks based on fundamental data. It takes more time building the dataset in Fidelity, which is quite extensive, than analyzing the data in STATA.
 
>You vastly overestimate the capabilities of someone that has yet to create a profitable trading system.

Not true. There is no possible way I have overestimated your ability lol.

>WL is also the only platform in existence with a portfolio level simulator advanced enough to not require vast modification to calculate simulations.

Absolutely untrue. And the other platforms you mention are also retail level platforms.

>I'll show you a quantitative analysis model I use on occassion to calculate the fair values of stocks based on fundamental data. It takes more time building the dataset in Fidelity, which is quite extensive, than analyzing the data in STATA.

Hate to break this to you, but we've done this work too... on as reported financials (you probably don't have access to those) on a basket of over 500 names going back as far as we could get clean data. It was a massive undertaking, that frankly wasn't worth it. Fair value might matter when you're trying as an RIA to sell a stock to a client, but we know that really has little to do with how stocks are priced.

I assume you tested this with something like a set of walk forward regression models correcting for common attributes like company size and sector bias? You can't BS this one because either you did the work or you didn't and either you did it correctly or you didn't. What fundamental factors did you find significant and how stable were the models over time? Also, what factors did you test and did you break out results by sectors and overall economic environment?

If you can give good (non-fumbling) answers I'll be happy to share some of our work on this topic. By trading this model in a very strange way we were able to build a portfolio that rallied off the market lows very close to 300%, but you could never have sold it to clients.

Finish one discussion first, but you keep opening cans of worms lol. Your answers to this will have to be good to get the goodies though... tell me you buy high ROE and low PB stocks and we're done before we start on this. :)
 
Quote from bwolinsky:


Sorry for the slow turn around. It's coming, and, Gann, I've already told you I can't analyze your data without it being placed into a spreadsheet. If Talon doesn't want me to use Excel, I'll use STATA 9 Intercooled.
More...

1) Open the file in notepad.
2) Cntrl-A to select all
3) Cntrl-C to copy all
4) Go to Excel and open a new file (Cntrl-N btw)
5) Select the first cell (A1) only and press Cntrl-V

Here endeth the lesson.

What do they teach you in finance classes in Kentucky? Excel knowledge is also table stakes for any undergrad hoping to work in the field... or at least it is in this city...
 
Quote from talontrading:

1) Open the file in notepad.
2) Cntrl-A to select all
3) Cntrl-C to copy all
4) Go to Excel and open a new file (Cntrl-N btw)
5) Select the first cell (A1) only and press Cntrl-V

Here endeth the lesson.

What do they teach you in finance classes in Kentucky? Excel knowledge is also table stakes for any undergrad hoping to work in the field... or at least it is in this city...
More...

If you attempt to insult my intelligence again, we won't be talking any further.

Enclosed is the spreadsheet when you paste everything into excel. I'm not going to be taking the time to compile it. Separated by spaces still goes horizontally, and there are A-ZZ spaces available. Everthing gets pasted into cell A1, and does not separate.
 

Attachments

Quote from bwolinsky:

If you attempt to insult my intelligence again, we won't be talking any further.

Enclosed is the spreadsheet when you paste everything into excel. I'm not going to be taking the time to compile it. Separated by spaces still goes horizontally, and there are A-ZZ spaces available. Everthing gets pasted into cell A1, and does not separate.
More...

From notepad to excel, every new row gets pasted horizontally in the first column. Again, also not something I'm going to work with.
 
Well... seriously though... getting a text file into excel is something worth learning to do. it is possible you have some non-standard settings selected on excel or something.

you can also explore the text import wizard.

didn't mean to be insulting but i dont know how you can claim to be the best system developer in the world if you can't get data from text file into excel. that doesnt seem right to me but i dunno

Quote from bwolinsky:

From notepad to excel, every new row gets pasted horizontally in the first column. Again, also not something I'm going to work with.
More...
 
Well, Talon, you are correct that I did get different values from STATA...significantly different actually, and I'll try to be aware of that in the future.

From STATA:

stats system2 system1 pairs

kurtosis 4.901559 25.44265 17.9387
skewness 1.030709 3.985537 2.377916
Median 0.0056425 -0.0155 0.007787
mean 0.0073903 0.0279479 0.0117409
range 0.0933859 3.778 0.2093699
max 0.0665603 2.896 0.1519635
min -0.0268256 -0.882 -0.0574064

From Excel:
System2 System1 Pairs
average 0.007149457 0.027947917 0.011740944
kurt 1.276175557 23.72214953 15.04170834
skew 0.867933014 4.049081018 2.382459037
MAX 0.064417476 2.896 0.1519635
MIN -0.026825569 -0.882 -0.0574064
Win 0.683673469 0.447916667 0.923761118

. sum Pairs, det

Pairs

Percentiles Smallest
1% (0.02486) (0.05741)
5% (0.00273) (0.05653)
10% 0.00084 (0.05649) Obs 787.00000
25% 0.00320 (0.04406) Sum of Wgt. 787.00000

50% 0.00779 Mean 0.01174
Largest Std. Dev. 0.01710
75% 0.01630 0.09834
90% 0.02819 0.12749 Variance 0.00029
95% 0.04066 0.13306 Skewness 2.37792
99% 0.07062 0.15196 Kurtosis 17.93870

. sum System1, det

System1

Percentiles Smallest
1% (0.88200) (0.88200)
5% (0.45200) (0.86400)
10% (0.21600) (0.79600) Obs 96.00000
25% (0.06800) (0.47100) Sum of Wgt. 96.00000

50% (0.01550) Mean 0.02795
Largest Std. Dev. 0.43706
75% 0.06300 0.31700
90% 0.17800 1.64500 Variance 0.19102
95% 0.23600 1.84200 Skewness 3.98554
99% 2.89600 2.89600 Kurtosis 25.44265

. sum S1732, det

<i>Thanks for teaching me the command.</i>

S1732

Percentiles Smallest
1% (0.01773) (0.02683)
5% (0.01138) (0.02375)
10% (0.00779) (0.01963) Obs 492.00000
25% (0.00197) (0.01853) Sum of Wgt. 492.00000

50% 0.00564 Mean 0.00739
Largest Std. Dev. 0.01402
75% 0.01402 0.05437
90% 0.02605 0.06442 Variance 0.00020
95% 0.03335 0.06624 Skewness 1.03071
99% 0.05201 0.06656 Kurtosis 4.90156

. sum B, det

BoWo

Percentiles Smallest
1% (0.06510) (0.06510)
5% (0.05140) (0.05510)
10% (0.04670) (0.05250) Obs 88.00000
25% 0.00035 (0.05180) Sum of Wgt. 88.00000

50% 0.01710 Mean 0.02575
Largest Std. Dev. 0.05767
75% 0.04670 0.15870
90% 0.15810 0.15880 Variance 0.00333
95% 0.15870 0.15880 Skewness 0.91114
99% 0.16090 0.16090 Kurtosis 3.63987


Interesting, to be sure. So, now that I've been through all the values and we agree on the calulcation, and you've heard what I have to say about system1 and your pairs model, I'll talk about system 2.

So your best system you consider with a 68% win percentage? I'm surprised you didn't pick the pairs model.

System2 is the least peaked out of all of them, and it appears the closest to a normal distribution. I do see some positive skewness. It appears it's the best because it doesn't lose as much when it loses? I see it having the highest minimum value out of all of your systems. Again, no matter which stat I look at, I resoundingly would choose the pairs model over the other two, but that's just me if I had to pick your "best system" from the distribution. Without seeing an equity curve, I can't say much. The system2 system appears to have the most "controll" over it's risk. Thinking it's a possible intraday system with the option to hold over-night.

I wish I would have looked at the summary page immediately rather than put you guys off, but I thought, as per a previous thread, you would want me to do the work, and in doing so, realized excel's calculations can be a bit questionable.

Thank you, Talon. The distribution and graphs would have been the easiest to work with. I think it's very obvious from your graph of my PTQQ2 system, that there is a lot of positve skewness and "fat tails" on the positive side.

System1's median is the lowest, and it's average is the highest. Mostly this is due to the win percentage. I believe you said this was the actual distribution from the S&P Add/Drop system, and I don't find it particularly attractive from the win percentage. I can appreciate a system that let's winners run, but if it's also doing so by letting a loss run to 88% down, I don't believe that's a very good trade-off. If the winpercentage was above 55-60%, I might feel differently about that possibility.

Your pairs model has "a lot of small profits", but you need a more concise ability to predict market direction, and the way I do that is by various lengths of linear regression predicted values and by testing statistical significance. It's not that those values predict anything, but that they provide direction. The win percentage is more from a lack of specific criteria for the entry, leading to marginally small profits. It's still what I would consider your best system, but I can see where System 2 has your best performance. I was thinking you considered system2 your best due to a high number of trades, but I see it has the least of all. Are these systems all over the same time period? I think system2 might have a shorter timeframe than the others.

So there's what I think, and I'm heading out for lunch.

I apologize for not saving us some time, but really, I just wanted to talk about this on my day off.--

Good Trading,

Good Luck,

Beau Wolinsky
 
Status
Not open for further replies.
Back
Top