HiI'm new here and I wonder if there's any compilation of the performance (and other indicators such as drawdown, risk etc.), by month or year, of all the screens. Much like AAII does for their own screens:http://www.aaii.com/stock-screens/riskreturnI've tried to find this information but it's not so easy for someone new at the forum. Thanks in advance.Nuno
People set forth performance for weekly, monthly, quarterly, annual screens; for #1 stock, 3, 5, 10, 20 or more stocks deep; with draw-downs, recent performance, standard deviation, sharpe ratios, performance during bear markets; people break down valueline-based screens, IBD-based screens, AAII's SI Pro-based screens, and gtr-based screens, not to mention screens taken from other sources. You name it and someone sets it forth.You can search for the words "Gold" and "Silver" and get Zee's analysis.You can program Robbie's gtr tester for some and test them yourself.You use Jamie's backtest.org tester for monthly or longer trading of many valueline screens.You can search for the word "torpedo" to see screens examined and then rejected despite initial backtest results.You can use Keelix' tester and search mode to see SI Pro screens.There isn't much agreement on "best" unless you specify your goal and trading frequency, plus source data you like or trust.Many people like YieldEarnYear, one of the "safer" screens for mid and longer-term holds.Adding poor recent perfomance to many screens improves results for very short-term holds.
The estimable Bill2m runs and posts weekly, an actual performance compilation for every screen since nearly the dawn of MI.The summary, is YTD and trailing twelve months, for 3, 5 and 10 stock versions.He posts a (massive) spreadsheet, updated weekly, which contains the actual results for every week for every screen going back to inception or 2001. It assumes a weekly trade.This is an invaluable resource - and he does not get enough recognition for this crucial and ongoing contribution to this board.http://boards.fool.com/Message.aspx?mid=30524967&sort=us...FC
I could not agree more. Bill2m does the board a wonderful service.
Thanks a lot for the answers. The table is indeed an extremely valuable resource. Let's see if I can make some money out of all this information... :)
The estimable Bill2m runs and posts weekly, an actual performance compilation for every screen since nearly the dawn of MI.These results are particularly notable because they cover each screenonly during the period AFTER it was created and started being posted.Thus, there is no hindsight or overtuning in the results.On the other hand they are based on weekly trading, which is far morefrequently than most MI practitioners follow because the trading costscan add up to so much during the year that it's hard for any strategyto be reliably outperforming to overcome that.So if you're not going to trade weekly (I sure don't), the relevance is iffy.If you are, don't forget to look up the rates of turnover and applya reasonably generous estimate for trading friction.Jim
Jim and others,You answered a question I had, but also raised one. Since Bill2m's spread sheet is for a weekly hold, can we assume that the a screen that does well on a weekly hold, would also do well on a monthly hold?Mark
Since Bill2m's spread sheet is for a weekly hold, can we assume that the a screen that does well on a weekly hold, would also do well on a monthly hold?Don't assume anything. These reports are useful for no more than identifying candidates for research. I would never pick a screen for actual investing from this spreadsheet. If you want to decide on a screen, test it in GTR1 for at least the last 15 years, with all-start-dates. Compare it to as many other candidate screens as you can, and then make your decision.Elan
Since Bill2m's spread sheet is for a weekly hold, can we assume that the a screen that does well on a weekly hold, would also do well on a monthly hold?Unfortuately, no.I've been meaning to post this for a couple of weeks now; this question prompted me to get off my duff and post it.I created a TVALUE screen on stockscreen123, as mentioned here: http://www.backtest.org/TVALUE:0612SBeaqG5XcpeG0XcpeB25Xr26T...A 4-week hold averages about 15% CAGR. Depends on the starting week of the rolling 4-week portfolio, ranges from 13% to 22%. Those of us who use SS123 have been aware that the backtested return can vary greatly with different starting weeks.So we generally do a 1-week holding period backtest and figure that this is probably the "true" return.But a one week hold of TVALUE is 9.9% CAGR.Basically, you cannot assume that a 1-week period and a 4-week period have similar returns. Maybe they do, maybe they don't, but you can't assume they do.FWIW, the other screens I run on SS123 do not do this. The others have the 1-week return solidly in the range of the rolling 4-week returns.
Elan,Thanks for your reply's. I have learned a lot from you. Your the one who taught me about keeping transaction fees to 1% of assets.
Thanks Ray.That is good to know. It would be nice if they correlated perfectly, but it is better to know the truth, than to lose money.
I agree with Elan in the assertion that you shouldn't make any general assumptions concerning screens based on bill2m's spreadsheet. As others have noted, a screen that does well as weekly hold, may not do well for other holding periods. As well, screen performance over time will vary depending on start date.Elan's real point is that you can't make long term inferences about screen performance from bill2m's report of recent screen performance. That's why he recommends long term back-tests.Some of the assumptions underlying each screen reflect investor's preferences. In the short run, the market may prefer one set of stock characteristics over another (value, growth, cyclical, momentum, etc). Bill2m's spreadsheet provides insight into those short term preferences.Some members of this board try to determine what stock selection methods appear most effective now. They like to adjust their screen selection to suit current preferences as reflected in the spreadsheets. Use care over time, as those preferences (frequently dictated by institutional traders) can change quite rapidly.Using this method will require considerably more effort and may not produce superior results over choosing one or more screens that have performed reliably over extended time periods.
I use the spreadsheet in two ways. One, I am one of the few who does have a portion of my MI allocation assigned to "what's working lately". Other portions are assigned to reliably backtested and high quality post-discovery screens - of which there are a very manageable number to choose from. What's working lately is not that much "considerably more effort" to figure out. You set up a spreadsheet with the return and deviation comparison metrics and rankings columns of your choice, copy and paste in the 5 or 10 stock values periodically, review the rankings and decide.Realizing that this approach is an experiment in style-chasing, it's not something I would recommend using for an entire portfolio. But I argue there's a place for it, instead of stubbornly sticking with a screen or an approach with a good backtest that has very visibly underperformed for some time.The second way is as a post-discovery or backtest validation tool. A screen that pops to the top of those rankings that has a low quality history can quickly be checked for reliability or deviation from normal. (like Beta or Option_A). Also, one could analyze the consistency of screen performance rankings over time. FC
I agree with Elan in the assertion that you shouldn't make any general assumptions concerning screens based on bill2m's spreadsheet.Bill2m has been posting the the results of weekly trading MI screens for a number of years. Quite an accomplishment.I wanted to see how the best performing VL screens have done versus the worst. The test used the Trailing Twelve Month results for consistency, and looked at 5-stock monthly trading. Only long screens were considered and only those whose results could be found at gritton's backtester.Here are the results for 2003 through 2011: Best 5 Worst 5 SPY2003 24% 58% 29%2004 14 33 92005 44 14 32006 12 20 142007 15 - 4 52008 -62 -54 -382009 31 49 322010 31 12 152011 -23 - 4 2.CAGR 3% 8% 6%DB2
Best 5 Worst 5 SPY2003 24% 58% 29%2004 14 33 92005 44 14 32006 12 20 142007 15 - 4 52008 -62 -54 -382009 31 49 322010 31 12 152011 -23 - 4 2.CAGR 3% 8% 6%
In the early days of MI, as it grew out of the Foolish Workshop, I'd been an avid follower/participant of the MI board. Then, I foolishly left to work on the other side (Salomon Smith Barney) for a time. In the past dozen years, or so, I've only followed the MI board sporadically.It had appeared that WWN or WWL had fallen out of favor. As an early proponent of "What's Working Now" (or Lately), I'm happy to learn I'm not alone in the "style-chasing" experiment.I still feel it takes more effort than choosing a screen or screens and just running the "rinse and repeat cycle". I wonder how many Fools do that? I suppose it depends on how involved in the selection process you want to be.Forever Foolish!
[highlighting largest number] Best 5 Worst 5 SPY2003 24% 58% 29%2004 14 33 92005 44 14 32006 12 20 142007 15 - 4 52008 -62 -54 -382009 31 49 322010 31 12 152011 -23 - 4 2.CAGR 3% 8% 6%Is that table right?Shouldn't the values in the first column always be higher than the values in the second?Maybe I misunderstood--was this the performance of the ones that had the highest performance the year before or simply highest performance in the year shown?I'm not sure whether this was intended to be a demonstration of howVL screens have done or how a VL screen WWL scheme would have done.Jim
It had appeared that WWN or WWL had fallen out of favor.Not for everyone. I have been continuously monitoring and improving a versionof WWL for several years now. I have a genetic optimizer that continually comparesweighting factors to see what performance you would have achieved from 1999 topresent using different strategies of WWL to select the screens to invest on basedon past CAGR and GSD to that point in time. The optimizer gives different resultsdepending on what you state as your goals (Max CAGR, Sharp, Min Ulcer index orweighted combinations of the factors). For most people minimizing a wild ride with aminimum Ulcer index would have been achieved by picking screens by their CAGR/(GSD^1.6).Better overall performance can be achieved with a wilder ride by choosing screens withthe best combination of long term and 9 month performance.RAM
It had appeared that WWN or WWL had fallen out of favor. WWL is what all those mutual fund chasers do. You know, the ones who show up in the surveys that show FUND performance is better than INVESTOR performance, since the investors continually jump around chasing the thing that worked best in the immediate past.Statistically, you have to go with the base rate.
Best 5 Worst 5 SPY2003 24% 58% 29%2004 14 33 92005 44 14 32006 12 20 142007 15 - 4 52008 -62 -54 -382009 31 49 322010 31 12 152011 -23 - 4 2.CAGR 3% 8% 6%Is that table right?Shouldn't the values in the first column always be higher than the values in the second?Dr. Pangloss would hope so, but in the real VL world the best performers from Bill's postings did not do as well as the screens at the bottom of the list.DB2
RAM,Would you be willing to post year-by-year results of the CAGR/(GSD^1.6) method and the combination of long term and 9 month performance method?Regards,Todd
Here are results obtained by starting with a list of 28 screens plus cash which changed to 26screens plus cash when short interest ratio became unavailable. The algorithm chooses screenseach month using only history to the selection time. No results are available the first yearuntil as there was no history to base selection on. One limitation of this logic is of course manyscreens used were not developed until the '06-'07 time frame. I can obviously get better resultstoday by simply choosing the X best screens to date, but those screens were not necessarilythe obvious screens to pick for the early years. My objective was to develop a methodology forchoosing screens that would improve as more data became available.The simple 9 month average gain/9 m GSD + Total long term average gain/GSDHolding Screens ranked 2:6 (1:5 works almost as well but slightly higher GSD)All screens have a 3mDDV of $600K which reduces gain but assures tradability.45 CAGR, 26 GSD, 1.56 RRR4, 23 Ulcer, 34 MaxDD1998-10m 28.801999 105.852000 70.602001 68.042002 29.522003 102.302004 81.642005 68.562006 59.592007 26.732008 -21.972009 32.442010 67.792011 -9.332012 31.60RAM
Shouldn't the values in the first column always be higher than the values in the second?Dr. Pangloss would hope so, but in the real VL world the best performers from Bill's postings did not do as well as the screens at the bottom of the listClearly we are talking at cross purposes, and/or I'm being thick.What is this table?If the first column is the performance of the best 5 VL screens in a given period,and the second column is the performance of the worst 5 VL screens in the same period,doesn't the first column value in every row have to be bigger than the second column number in that same row?i.e., if the best 5 screens returned 31% in 2009, then how can the worst 5 have returned 49%?Wouldn't the worst five be, err, worse?You've ranked them on the 12 month performance, right?Or something else? Trailing 12 month performance?Clueless,Jim
Clearly we are talking at cross purposes...What is this table?Since 2003 Bill has made available a ranking of screens by trailing twelve-month performance (based on weekly holds, IIRC). At the beginning of each month I noted the five best VL screens (5 stocks) and ran those through Jamie's backester to get results for that month. This chart shows the results by year for investing in the current five best performing VL screens.I also repeated the procedure for the five worst performing screens, TTM, each month. The five worst slightly outperformed the five best. Neither group shone compared to SPY. To me this indicates that using Bill's rankings does not add value and suggests the general use of VL screens does not add value (see also the results in the MIVL Index thread).DB2
Clearly we are talking at cross purposes...What is this table?Since 2003 Bill has made available a ranking of screens....Ahhh!OK, it's a monthly WWL test sorted on the trailing year weekly performance results.I didn't get that, I thought it was just a report on the VL screen performance.Never mind, I'll go back to sleep.FWIW in the past I found a (smoothed) 3 month lookback WWL added value, but not much.I still kinda like my brain-dead method for picking screens:Find out which long screens do best in bull markets and which do best in bear markets.Then forget about timing.Throw out the bull market ones and use the bear market ones all the time.If the bad times are OK, the good times will take care of themselves.Jim
Dr Bob's results of choosing VL screens based on past 12 mo performance doesn't come upwith the same results I do using a smaller set of SIP screens. I find that the past 12 monthsperformance has some predictive value for next Month's gain. But only about a 7 to 10%advantage over picking randomly.I should also point out that my posts on the advantages of using a 9 Month WWL was basedon choosing from a list of screens that had already been chosen as much better than averagescreens to start with.(although once chosen I kept them in the selection pool even if they tanked post discovery) They had all continued to have value with at least a hold 10, and 2 monthperiod, better than average Sharpe. In fact a random selection from those screens would have given a CAGR of 35. So using a selection based on a combination of long term CAGR, GSD and9 Month CAGR and GSD in this case seems to add another 5 to 10% and reduce the GSD a little.The point is that I think 90% of the effort should be in the initial selection of screens.For the same rational Zee's SIP based gold screens continue to hold up post discovery.RAM
Three general observations concerning DrBobZ comments.First, my initial experiments with What Works Now were focused on trading with weekly holds. My focus is still weekly.Second, I always have employed some kind of smoothing over several weeks reported performance. The time periods I've used have ranged from 5 to 13 weeks. The object is to discover some consistent screen performance over the recent short run. Unfortunately, I currently have no scientific evidence supporting any smoothing method or the number of weeks to use. Maybe I've too many other interests or become lazy and less rigorous as the years have passed.Third, just like most other MI methods, this approach doesn't seem to have any predictive power when all hell breaks loose, as in the unpleasantness of 2008, etc.Finally a casual observation that has probably been discussed and beaten to death here. The SIPRO screens seem to do better than the VL screens. More to the point, most of the selections from my recent WWN efforts have come from SIPRO screens and seem to do well.That and 2 bucks will get you some java at Starbucks. Unfortunately, inflation killed the 2-bit cup of coffee long ago.
Best Of |
Favorites & Replies |
Start a New Board |
My Fool |