Rick67
64 posts
msg #101082
- Ignore Rick67 |
6/5/2011 10:15:56 AM
Great work Kevin...
RE: "So in a single massive optimization"
You have a very smooth equity line over a long period of time which is always a great sign.
Just want to throw this out there:
Optimization can be a trap since your are "curve fitting". It looks great on paper, but next month, the market's "curve" could be different and your system could fail.
Have you tried doing walk forward testing? Or testing different periods over the past as opposed to one large period?
I do appreciate your postings and like to follow what you do!!
|
seanban
22 posts
msg #101083
- Ignore seanban |
6/5/2011 12:05:32 PM
Kevin,
If your filter run example, you posted the results returned last Friday and selected just 3 stocks: AMD, SWY and DD. Is there some reason that the other 11 were not seleted?
Thanks - Sean
|
Kevin_in_GA
4,599 posts
msg #101085
- Ignore Kevin_in_GA
modified |
6/5/2011 12:36:21 PM
Rick:
Of course. Notice that I used the period from 12/31/1999 until 12/31/2009? That was so that I could use 1/1/2010 until today as my walk forward period. The goal is not so much to beat the market by the same percentage, but to keep a high Sharpe ratio and profitability.
For the period 12/31/2009 until 5/31/2011 the system gave a CAGR of 16.14% versus 14.6% for the ^SPX, a return on starting equity of 23.6%, and a Sharpe ratio of 1.56 (annualized). Average days held was 5, average return per trade was lower (0.62%, not nearly as good). However, you were only in the market an average of 53.40%, so your money was sitting safely in cash (or out of play, depending on your mindset).
Now one of the nice things about Stratasearch is that it can automatically update the filter as time moves forward to give you the "current" best settings, based on whatever look back period you choose. This keeps the filter updated for the most profitable settings as time progresses. Over the last 2 year look back period, the optimal settings have moved a bit ( now are at 15 days, -2 entry to 0 exit, williams %R(15) > -89, close above ma(150)).
This is an interesting capability (having your filter adapt and update, say every month). I have not used it or backtested to see if doing so would have been a better approach. I'm just not sure that this isn't curve fitting carried to the next level.
|
Kevin_in_GA
4,599 posts
msg #101086
- Ignore Kevin_in_GA |
6/5/2011 12:39:26 PM
Sean:
That is not my example - StockFetcher just shows you three of the returned stocks. You need to click on the filter and pick the top ranked stocks (I have them ranked by lowest Z-score16 at the top).
|
seanban
22 posts
msg #101089
- Ignore seanban |
6/5/2011 2:45:16 PM
Kevin,
I must be making a mistake somewhere; when I click on the filter indicated in this thread (msg #101013), I get 11 symbols listed, three of which are those listed by you. See below.
Am I missing some extra filter here?
Thanks - Sean
FMC 79.68 -2.16 682100 -2.86
LEG 24.27 -2.53 1434500 -2.62
AMD 7.97 -3.28 21264700 -2.49
SAI 16.76 -2.61 3422400 -2.34
SWY 22.28 -1.76 8320900 -2.27
SNA 57.07 -1.86 339500 -2.26
PKI 26.4 -2.19 810400 -2.25
ESRX 56.73 -2.26 4701700 -2.23
TER 14.93 -3.37 4014000 -2.19
DD 50.29 -1.72 6056500 -2.12
COST 77.81 -1.51 3377400 -2.11
WU 19.8 -2.7 4911500 -2.11
MOLX 25.92 -2.7 720800 -2.03
PDCO 32.77 -1.77 895200 -2.01
|
Kevin_in_GA
4,599 posts
msg #101091
- Ignore Kevin_in_GA |
6/5/2011 4:30:05 PM
If you want to paper trade this, you are going to need to set up a separate filter to tell you when to exit.
Example: My entry filter is this:
From Friday's data I choose the top 10 stocks with the lowest zscore16 values. They are:
FMC 79.68 -2.86
LEG 24.27 -2.62
AMD 7.97 -2.49
SAI 16.76 -2.34
SWY 22.28 -2.27
SNA 57.07 -2.26
PKI 26.4 -2.25
ESRX 56.73 -2.23
TER 14.93 -2.19
DD 50.29 -2.12
OK. Now I create a watchlist with these stocks in them, called "zscore_portfolio", put in my entry prices. I would run this filter toward the end of each day (after 3:30 PM, sunce the stocks are less volatile then), so you could just use the close and update any paper trades with EOD data.
Now to make sure you know when to exit (since we exit at zscore(16) > -1) you simply use the following filter:
This filter won't work here on the forum because there is no such watchlist, but it will work on your machine. Each night have both filters emailed to you. When one of your current holds crosses above -1, it will show up in the second filter - sell it and buy the stock from the first filter with the lowest zscore. Remove the sold stock from your watchlist and replace it with the new pick.
Kevin
|
novacane32000
331 posts
msg #101092
- Ignore novacane32000
modified |
6/5/2011 4:59:22 PM
Kevin
I used a 7% stop loss along with your zscore exit and had excellent backtest results dating back to 2002 .
I also through in exit after 5 days and improved results further.
I do not know much about curve fitting-maybe I am guilty of that here.
Hope this link works to the backtest results.
https://spreadsheets.google.com/spreadsheet/ccc?key=0Aunlddkvv9P5dHJuLXJ2UkVuaGJzYmFNdWh2OGIycHc&hl=en_US#gid=0
|
seanban
22 posts
msg #101120
- Ignore seanban |
6/7/2011 4:44:46 PM
Kevin,
Started paper tradin as of yesterday based on both filters provided earlier (entry and exit). What I noticed is that during certain times of the day, exit signal list varies. At EOD, we have one signal - DD. But earlier, there were several. Just wondering, performance wise, is it best to exit when the signal is given or to wait as suggested until 30 mins. before markets close?
Thanks - Sean
|
Kevin_in_GA
4,599 posts
msg #101123
- Ignore Kevin_in_GA |
6/7/2011 9:06:21 PM
I see two signals today - DD and ESRX. I did not check intraday. Both closed at modest profits.
Four new picks were signaled from the entry filter. The top two are Lorillard (LO) and Ryder (R). I would replace the closed trades with these using the open price tomorrow morning.
Let's see how this plays out ...
Kevin
|
calhawk01
135 posts
msg #101125
- Ignore calhawk01 |
6/7/2011 9:36:26 PM
Kevin, I just want to say that you are awesome! Thanks for sharing your ideas and thoughts. Very helpful!
|