We’ve launched Canada and the new USA data. Please read the document “Canadian Data” in our Help->Resources->About our Data section. One important section is about the differences with USA & CAN holidays. In short, your simulations will adhere to the holiday schedule of the selected benchmark. Therefore we recommend using Canadian benchmark (only one at the moment) when using the Canadian universe, and vice-versa.
This new release also has a rebuilt USA data that has been discussed in previous posts. This will affect your re-runs of simulations, but not substantially. We’re making available the old USA data in the beta server at http://65.84.1.52/index.jsp if you need to compare old vs. new.
Other key points about Canadian data:
You must use the universe ‘All Fundamentals - CANADA’
Lots of penny stocks in Canada, be sure to use something like Close(0)>1
We only have one Canadian benchmark at the moment
Canadian data is NOT the same. A lot of companies have interim periods with missing data, so TTM ratios can be N/A even for things like SalesTTM
Great thanks. If I want to publish an R2G, can you please allow $TSX as an hedge? Just force model revision whenever you have XIU. I can build you an extended 1999-2014 XIU if you want. I have a $TSX total return.
I checked two more sims, one slightly off, another total return reduced about 10%. Did you remove stocks from Universes like PRussell, and AllStocks?
Maybe you should back off all your changes until you find the bugs.
It is possible this is creating biases like “always buy Berkshire Hathaway (BRKB)”, because it would produce a very low price/projected earnings number as the EPS would be based on class A shares. As it happens that particular stock has done quite well over the last few years, so it has been a good pick, but for the wrong reason.
In this regard, this is a major pitfall of backtesting - nevertheless, the reduced return is a more realistic approximation, and one also has to ask the question, why the change is so much? If the simulation has a low number of total trades, this implies that it is not so robust. Remember there is a lot of noise in the stock market so be careful about curve fitting (fitting to noise).
sglinski, please see the two sims with DEBUG in the name. One was run with the old USA and tot return is 59% the other with new USA is 45%.
Note: It’s a 4 stock sim ran for 5 years. Any single change in the picks could account for the difference
So I zeroed in on the first difference in 7/31/09: sim with new data buys PCZ^09 (Petr-Canada) the other doesn’t. It boiled down to different PEG because the estimate is coming in Canadian dollars. As part of the rebuild we fixed the problem that Canadian companies trading in USA (therefore USD prices) with Canadian estimates were not being converted to the same currency as the stock.
Looks like we need to rebuild our Prussell universes since they should not include Canadian companies. A simple test with Country(“CAN”) with Prussell 3000 returns about 100 companies.
Thanks for the quick reply-
So, in the old sim, a dollar of Canadian earnings was reported as a US dollar of earnings, inflating earnings about 10% based on the current exchange rate?
Or, was it a dollar of Canadian stock was reported as a US dollar of stock, inflating the share price?
The former would seem to overweight Canadian companies ranked by lower PE, the latter underweight them.
Maybe I could recreate the old result by artificially weighting Canadian stocks?
As far as the position size, I have found that for the vast majority of my sims, the postion size that optimizes the sharpe ratio is 5 stocks. I’m not sure of the reason for this, but perhaps it relates to industry/sector cross correlations. Anyway, I compensate for resulting higher variability of future returns by combining ports (or book use) and sizing positions relative to my total portfolio, a la Van Tharp etc.
" the postion size that optimizes the sharpe ratio is 5 stocks"
This is absolutely true. Several years ago I did some testing on individual ranking factors and systems while varying numbers of stocks. The test was a relative sharpe ratio i.e. relative to the stock universe. Five stocks was the magic number for optimal Sharpe. The reason is that if you hold too few stocks then there is too much volatility. As you increase the number of stocks the performance goes down.
But this is in backtest. Going forward, you can expect more volatility and hence the number of stocks needs to increase, either by multiple systems or increasing the number of holdings for one system.
Not exactly . The NextYEPS in the old USA was 4.5 CAD, the new one is 3.9, so the PEG is now higher in the new USA. However the NextYEPS is now lower than trailing TTM which makes the growth N/A. In these cases the algorithm “falls back” to a PEG calculation that uses analysts LTGrowth rather that evaluating to NA.
PS: I would not go about adjusting your system. That’s exactly why we almost always see lower returns with slightly different data points. Systems are being constantly tweaked until the max returns are achieved… the definition of curve-fitting.
Hi Marco,
For existing ports and sims do we have to add Country(“CAN”) = False or the Canadian stocks are only selected if that universe is specifically enabled.
Thanks.
KJ