| StockFetcher Forums · General Discussion · Technique: Calc your own returns - for any period | << 1 2 >>Post Follow-up |
| cegis 235 posts msg #33296 - Ignore cegis |
9/17/2004 5:23:18 PM This is a discussion of a technique that I developed a few months back, and has seemed to be useful regularly in other threads ever since. So, I decided to document it, in the hope it will be of use to others. This technique allows you to calculate returns for a filter, for any time period, and (almost) any assumed trading "program". What StockFetcher offers as basic site functionality is a column labeled "Performance" when showing results from a filter that has the "offset" phrase. The performance shown (I think - correct me if I'm wrong) is based on the closing price of the stock on the day that the stock matched the filter (the offset date) and yesterday's close. This means that the period covered by the column changes with the offset. You can also get the 1 day, 2 day, 3 day, 4 day, one week, two week, etc. performance for the filter's results by clicking on the Performance column header, and "browsing through the dates" (assuming there have been that many trading days in the offset period). This is good, but limiting. The scenarios that SF's Performance functionality does not take into account are things like "performance for a 10 day period, regardless of the start date", "worst case performance" using the day-after-match's high and end-of-period low as entry/exit points, or "trading program rules" such as max loss of 2%. This technique allows such analysis. The "trick" to this technique is to use the "days ago" phrase instead of the "offset" phrase. This allows you to "see what happened" AFTER the stocks matched your filter. As an example, if you use "offset 10 days", SF will not allow you to retrieve ("see") values from what would be "offset 9 days". However, if you use "10 days ago", you can always analyze what happened "9 days ago". Let's look at a simple example. We'll assume our starting filter is: If we run this filter on Sat. 9/25/04 (which will be our assumed run date for all of the following discussion), we'll see the results for 9/24 market close. Now, let's assume we want to see how this filter performed based on picks from 10 days ago. If we use we'll get selections for 9/10, and we'll get the Performance column talked about above (9/10 to 9/24, close to close), as standard SF functionality. However, if we wanted to see what a "worst case" scenario would have been, we would not be able to access the data for 9/11's high or 9/24's low to make this determination. However, if we instead used we'll still get the results of our original filter as of 9/10, but now we can calculate our worst case scenario: Since all of our conditions have "10 days ago", the first set{} command is "looking into the future" to get the following day's high (high 9 days ago), and the 10th day's low (low [0 days ago]), to calculate the return ($). The other two set{}s just convert the first set{} into a percentage. (I'm using my entry as the basis for the percentage...) The add columns will show us our result. Note that the return calculated will ALWAYS be a ten-day return. We can continue to backtest this filter by adding an "offset" too! With the offset 1, we'll actually match stocks for 9/9, and show what the 10 day return is through 9/23. You can then alter the offset to continue backtesting other dates. In other words, you do NOT have to keep changing the "days ago" in order to backtest prior dates. Just remember that SF will report "results as of 9/24" (or whatever your offset would suggest) at the top of the results list, and you have to remeber that you used "days ago" to shift the data back an additional 10 days. Also note that you can calcualte multiple returns for the period covered by the "days ago" offset: Here, we added a claculation of the 5 day worst case scenario, in addition to the 10 day return. This can be taken one step further (this is where this gets REALLY interesting!) by using "logical variables" to actually implement a "trading program". (Logical variables are discussed in several other threads, so I'm not gonna go into them here...) Let's say that you want a 2% trailing stop, and you'll hold for a max of 3 days. (I'll discuss this day limit later.) We'll assume we enter at the opening price, and exit at our stop or 3 day close. We'll also assume we only adjust our stop once a day based on the open. This filter will show you how well you would do: So here, out1, out2, out3, and out3c will show you your exit price if you exit on the first, second, or third days, or at the close of the third day (respectively). They will be ZERO if you did not exit on that day. Thus, you can see when you would have exited, as well as how well you would have done. There are two things I would have liked to do in the prior filter, but SF limits how many times you can "nest" set{} statements (that is, base one set{} on the results of another). I would have liked to change the calculation of limit2 and limit3 so that it only goes up in value (a trailing stop), as well as calcaulte the $ and percentage return based on my exit price. I would have also liked to check for "green hold" (high 2 days ago > close 3 day ago) before considering myself "in". However, this filter sits right at the edge of the SF nesting limit, so I was unable to do that. That's also why I chose a three day maximum hold period. One way to get around this, somewhat, is to show results for every day (instead of figuring out if you would have been stopped), like this: Now, you can read across the columns, and if you assume you use 2% below the high as the following day's stop, all you have to do is read across the columns to see where a "low" value is less than the prior column (profit or limit) to see where you would have been stopped out. This isn't perfect, but it can get you some useful information! Things to Consider ================== - The number of days you want to use in the "days ago" phrase is the maximum holding period (or days worth of data you need) to calculate your returns. You do not want to use a larger number. - If your filter compares to prior days (ie, something like "open is greater than close 1 day ago"), you want to ADD the number of days in this calculationally-based days ago to the condition's days ago (in this case, making it something like "open 10 days ago is greater than close 11 days ago"). - Note that the days ago prevents you from running the backtest as of a date "closer" to today than the number of days ago used. I do not consider this a major limiting factor. - SF will report "Filter results for <day of week> <date>". This will be based on the offset, not your days ago, value. You just need to remember that you have programmed an additional offset into the filter for backtesting purposes. - Whenever I use this technique, I first copy the filter that I want to backtest, then modify it to do the calculations and use the days ago that I need. You don't really want to use days ago (in the manner described here) in a filter you're gonna actually select stocks from... - Depending on your trading conditions and return calculations, you may be able to use the count() and sum() functions, and/or the "<n> days high" / "<n> days low", to look at longer time frames than the three days as in the example. - If you are backtesting a weekly filter, use "weeks ago" instead of "days ago". Also, funky things will happen if your filter mixes weekly and daily indicators. (Basically, the "days ago" and "offset" are not quite equivalent due to the way SF handles weekly values...) - Remember that you can download the results into a spreadsheet, and then use that to extend the backtest based on any logic you couldn't get into the filter. Hope this helps, C |
| Joules360 30 posts msg #33300 - Ignore Joules360 |
9/17/2004 11:56:53 PM Cegis: Wow!..You definitely know your stuff!...I apreciate your work...Stockfetcher should hire you as a consultant...Thanks for the help on my Le-Sling scan backtest as well. I haven't had time to read all the threads here..but is there a scan you prefer and have had good success with? I like my Le-Sling...but I am always open to another great scan for finding winners. My strategy is to find the meat of the run of a documented safe uptrending stock...NO pennies etc. I have my whole retirement that I'm controlling and the pennies just dont do it for me...hehe... Just looking for 10% hits over 10 days or so to avg with the small gainers and losers to attain that 1% week net gain. Doubling my money every 2 years is my goal. You probably have read the system of buying/selling I am using. I understand if you dont want to disclose it...but being that you are so proficient in this programming...I figured you must have backtested some that have great potential? Thanks! Joules360 |
| cegis 235 posts msg #33323 - Ignore cegis |
9/20/2004 8:48:37 AM Joules360, > I apreciate your work... I'm glad somebody does! <GRIN> > Stockfetcher should hire you as a consultant... I Agree!!! (I could use the work!) > Thanks for the help on my Le-Sling scan backtest as well. Glad I could help... > I haven't had time to read all the threads here..but is there a scan you prefer and have had good success with? To be honest, I've yet to find a filter that *consistently* produces more winners than losers. Most that I have backtested seem to follow the general market, more or less. Money management (therefore) is the key to producing an overall gain on a short-term basis (or any other term basis, for that matter<g>). Good luck! C |
| Joules360 30 posts msg #33328 - Ignore Joules360 |
9/20/2004 2:08:17 PM Cegis: I agree the general market is a factor in all of these scans...that is why I developed the opposite Le-Sling scan for shorts as well as well as the Long scan...This way...if the overall daily trend is down...you can go with the short picks...My window trade broker does not allow shorting...so I will have to deal with the longs only. Joules360 |
| kgriffen 49 posts msg #33339 - Ignore kgriffen |
9/21/2004 8:38:11 PM cegis, I am currently furiously trying your suggestions. They are great! How would you re-write this line for ten days ago? and slow stochastic(14) Fast %K 1-week low was less than 50 within the last 5 days |
| kgriffen 49 posts msg #33340 - Ignore kgriffen |
9/21/2004 9:03:46 PM Ok, I think I need help. I am getting set{} errors in debug. Maybe for ease I can post my filter here for suggestions: Here is my Mother of All Filters: Now, that gives me a list of SP500 stocks that have closed above their ema14 today with a white candle, and has Stoch and MACD features that I like. Tommorrow, I would only purchase the stock if it went at least .05 cents above today's high, confirming what I hope to be the start of a trend. So my one day backtest filter reads as so: Now, here comes the tricky part. I immediatly set a stop for 3% -OR- .01 below yesterday's low, whichever is less. I also will close the position if the close falls below the 14 day EMA at any time. With your new advice above, I would like to build a filter that tests backwards 5 days. I am having problems with these two lines: and I would also like to see if I got stopped out either by my stop or by the close below ema14. I tried to write it, but my brain is at its own set{} limit. Any help would be appreciated! I am currently seeing how Le-Sling may help this filter as well. :) -kg |
| cegis 235 posts msg #33343 - Ignore cegis |
9/22/2004 8:43:53 AM kg, First, there's an issue with the way you wrote these two lines in your backtest filter. The way you have it is not a "1 day ago" equivalent to the "normal" filter. What you want for these two lines is Note that when you are trying to "shift" an expression by a number of days, you need to shift *each indicator*, not the period of time covered by the phrase. Your "slow stoch... ...the last 5 days" and "count(... ..,10)" are both increasing the time period viewed by the filter - the condition has to be true for a longer period of time, and/or you're covering days not covered in the original filter. You want to add "1 day ago" (or "yesterday", as you point out) to the values being inspected, and leave the time period alone. Is this what you mean by "having problems with these two lines"? (For future reference, it helps *a lot* if you can be *very* specific as to what problems you seem to be having. Just friendly advice...) Oh, as for your first post's question, I'd write the line as: and slow stochastic(14) Fast %K 1-week low 10 days ago was less than 50 within the last 5 days You can test this by adding columns for "slow stochastic(14) Fast %K 1-week low 10 days ago" as well as "slow stochastic(14) Fast %K 10 days ago", 11 days ago, 12 days ago, 13 days ago, and 14 days ago, and manually verify that the 1-week low is the lowest of the individual values. One last thing: The filter debugger currently only allows 10 set{} statements before it shows an error message. The actual filter execution does not have such a limit, so the "limit exceeded" message can be ignored in this case. You know when you've hit the filter's nesting limit when it will no longer show the variable being set{} as a column in the results. If you are still having trouble, post the filter with the error and I can take a look at it. HTH, C |
| kgriffen 49 posts msg #33368 - Ignore kgriffen |
9/23/2004 10:44:00 PM cegis, very helpful as always! thanks, kg |
| cegis 235 posts msg #33370 - Ignore cegis |
9/24/2004 8:21:35 AM kgriffen, Glad I could help! Good luck! C |
| cegis 235 posts msg #34758 - Ignore cegis |
1/5/2005 11:57:01 AM People have been asking about this over the past few days, so I thought I'd bring it to the top of the list of threads... This allows you to backtest for any period, including some entry/exit strategies. HTH, C |
| StockFetcher Forums · General Discussion · Technique: Calc your own returns - for any period | << 1 2 >>Post Follow-up |
Copyright 2022 - Vestyl Software L.L.C.•Terms of Service | License | Questions or comments? Contact Us
EOD Data sources: DDFPlus & CSI Data
Quotes delayed during active market hours. Delay times are at least 15 mins for NASDAQ, 20 mins for NYSE and Amex. Delayed intraday data provided by DDFPlus