I am still in the "trial " phase of my RW expierience. Can anyone share their real world results using this tool?

Thanks,

Boyd


Attach file (image (.gif, or .jpg), screen (.und) or report (.rpd))
Robert
at 10/28/13 11:06 PM : RE: real world results


It can work pretty well, but there are some significant limits.  These range from specific constraints/problems to general rule of thumb guidelines.

Specific constraints:

1)  Do not use "index member" in backtests.  The listing is only for current makeup of a given index and will give you look forward biases.

2)  Do not constrain to COM or ADR in security type.  Also, do not use "not" OTC.  Once again, uses current (as in now, not at point in time when the backtest is  supposed to be working).  Could use "not" ADR, though, since nothing becomes an ADR after being COM or OTC.

3)  The returns in the backtest are total return numbers (div + price).  RW simply assumes that whatever the indicated dividend of the stock is at the time of trade entry will accrete linearly over the hold period.  While this is not much of a problem for long hold periods (24, 52 wk) it can be problematic for shorter periods.  Also, it does not work for ADRs since they do not have the same div payment characteristics as US stocks.  Also blows up if trying to grab stocks with high yields since it doesn't account for dividend cuts or suspensions during the hold period.  Can double check breakdown of a backtest returns by downloading backest detailed holdings (right click on the average label at the bottom of the output).

Rules of thumb:

1)  The more securities a backtest returns in a period, along with higher market cap, trading volume and price, the more likely the results will be something close to what you could have actually done at that point in the past.  I would be highly sceptical of the results from strategies that are at the other end of those ranges.

2)  The longer the hold period, the lower the transaction cost impact will be.  Do not kid yourself about this.  If you gin up a weekly screen that beats the SP500 by, say, 50% a year, it is almost certainly a mirage.  Where you actually execute versus what RW assumes will wipe out most if not all of the excess return.

3)  The backtest results are an approximation.  The DBCMHIST database from which they are generated is not absolutely "stable" over time.  Can't say exactly why this is, but it is clearly a factor when backtesting screens with more complicated criteria.  You will not be able to see this for yourself since you will not have a DBCMHIST from a year ago to run RW off of and compare the output for the overlap periods.  It is not a huge issue if you keep my comments in (1) in mind.

4)  Don't think that because a screen has great backtest results that it will continue.  Excess returns can disappear surprisingly abruptly.  I would focus on finding something that generates a rolling cumulative excess return that is as smooth as possible.  You will have to export the results to Excel to generate these plots.

5)  Realistically, a 5-10% per annum excess return is possible (but by no means a given).  If you gin up something that is getting some multiple of that, I would double check to make sure that entry / exit levels are accurate/realistic and that dividend assumptions aren't skewing it.  You almost certainly are not going to find some money printing machine that all of the quant funds out there have missed (they have a lot more resources than you or I do to throw at the project and they are not exactly setting the world on fire with their returns recently).

Hope this helps.


Attach file (image (.gif, or .jpg), screen (.und) or report (.rpd))

For questions or more information or to Order Today, call 1-800-767-3771, ext. 9392 or 1-312-630-9890, ext. 9392
You can also e-mail our Sales Support at: kevinm@zacks.com
Copyright 2012
Zacks Investment Research
111 N. Canal St., Suite 1101, Chicago, IL 60606
(800) 767-3771 ext. 9392