its not the sector. i think it’s not enough memory. it’s back now, but prob won’t last all night
Wow, I did not think Canada has over 7000 Stocks!
XX
You have to login in the beta server, the internet IP address in my first message
My mistake. I opened a second window. And that wasn’t in beta server. Sorry.
Seems ok. Don’t use any custom universes
Judge,
On the screener, Canada only shows 1812 stocks. With only 383 over $500k liquidity in past 50 days, and close>$2 and >$50MM mkt cap. Where are you seeing 7,000?
Best,
Tom
Marco,
Please correct me if I’m wrong. However, using the screener, I am not seeing any overlap in the Universes. Steve suggests there should be. Can you confirm if there are stocks that exist in both universes? If there are, can you give us a way to exclude them (a rule). Thanks.
Best,
Tom
It appears the “frank” does not work for Canada yet?
Seems to work for me. Which screen?
All pre-defined universes are only for US. You don’t need to specify Universe(AllFund)=0
All stock in CAN are Canadian stocks. The overlaps would be with companies that dual-listed: different tickers one in US one in Canada, like ABX & ABX:CN
We will add a factor like ‘DualListed’ that when run in US universe OR Canada universe will give you only the companies that are dual listed. Note that in most cases one stock will be very liquid, the other not so much, but arbitrage keeps performance the same.
Not sure I would call it “out-of-sample”. A lot less stocks makes the ranks much different. Also the sector weights are completely different. See our document in HELP->About Our Data
Marco,
You’re correct, frank is working. It appeared not to be working for me, because the liquidity distribution of the Canada universe is very different.
As far as out-of-sample, I am simply screening rank performance for top 30 and 50 stocks - using only a liquidity filter. If I am able to avoid all dual listed stocks, I am thereby getting a fair sense of how the ranking system did on a Universe of stocks that could not possibly have been part of the system development. This is equivalent to truly ‘held out’ data set. So, the insights at these levels are valuable. To me at least.
I agree it’s not ‘out of sample’ in that factor trends were likely very consistent across countries across this period. And market conditions (such as the proliferation of automated trading and backtesting platforms) were coming on-line in all countries, etc. Also agree if looking at only ‘rank’ bucket performance can be misleading. So…not sure what to call it. But…it is a ‘clean’ data set for the same period.
Best,
Tom
Thanks for the addition Marco, its nice to have, even though liquid issues are relatively limited (about 300 C>1&AvgDailytot>$1m) an many of those are dual listed in the US.
I think if you want to add markets UK or Japan would be more interesting as they are relatively big, less correlated, and have less overlap with US, but before that, personally my view is that there are more important things such as VWAP in simulations, hedging rebalancing, and of course, daily rebalancing. US markets are big enough, having others is certainly nice, but I would not call it a must.
I think it would be appropriate to have those markets as an extra service, like market data usually works for most providers, so only the people that use it pay for it, I am not sure if compustat or whoever the best data provider will be for foreign data would be willing, but an agreement of revenue sharing should be ideal to also minimise risk for p123.
Keep up the great work! p123 is amazing!
Steve said:
[quote]
As far as out-of-sample, I am simply screening rank performance for top 30 and 50 stocks - using only a liquidity filter. If I am able to avoid all dual listed stocks, I am thereby getting a fair sense of how the ranking system did on a Universe of stocks that could not possibly have been part of the system development. This is equivalent to truly ‘held out’ data set. So, the insights at these levels are valuable. To me at least.
[/quote] Steve,
This is why I always develop my Sims using EvenID = 1. That gives me an equal size universe of stocks to compare to with EvenID = 0. This way I am comparing the same market variations and I am sure that I haven’t cheery picked stocks during development.
Denny 
Denny,
That’s a great idea in theory (developing Rankings on Even/Odd). But…most long-time users of P123 know the major factors and tendencies of the US data backwards and forwards. New factor discovery happens. But is fairly rare. So…one major problem is that we basically know what’s worked. Or been written up in articles, etc. In a few months the same will be true of the Canadian data. But…right now, when it’s fresh and new…we have a chance to test systems on a large chunk of held out stocks. I’ve been doing that all morning. I’ve found confirmation of good results on a couple of factors that I was worried about…and found one other system that seems to have some minor issues against it. I’ve also been able to run A/B tests on multiple versions of rankings I had been debating between before. So…So far…it’s interesting and - feels like something that’s fairly rare for us as system builders. So…it’s worth thinking about how best to use that data…before we learn the tendencies of the Canadian market as well as the US data set.
Best,
Tom
Aurelian,
It seems to me there may be an error with your Canada tests. Or perhaps the P123 Canadian Universe has changed? Or your not using rank at all in sell rules?
After looking at several of my live money sims, I find that, at a minimum, if I use rank at all in the sell rules, the rank values frequently have to be adjusted because the universes are very different sizes and liquidity compositions. I find that I typically have to lower Rank for Canada if it’s used in a sell rule to keep annual turnover similar between the US and Canada.
I see that your turnover rates are very similar for US and Canada, but you’ve made no system changes? I assume rank is some part of your sell rules?
I also find that some common factors I use to try and trim potential ‘left tail’ stocks, end up not working in Canadian sim’s. They reduce the universe to very small sizes. And have to be deleted or have the values adjusted. This is simply due to the very different number of stocks available in a tradable liquidity range.
I also cannot construct custom universes for Canada. Even on the beta server. I can construct them, but get an error when I run the sim. And your sim’s use these. Hmmm…
So…I’m curious…Do you still get the same results this morning?
Best,
Tom
THANK YOU FOR THE CANADIAN DATA!
@Tom, I get the same results but haven’t tried to create more customs universe. I haven’t changed a thing, just copied the R2G port into a sim, went into custom universes and copied the R2G relevant universes into new ones to change no-OTC by All fund-Canada. That’s it. Didn’t touch the buy or sell rules. I know my universe shifted from 3500 stocks to 500 or so but somehow it didn’t affect the turnover too much.
Backtesting on Canada did make me revise “Aggressive Value” on R2G. So far the only liquidity rule was mktcap>100 but the sim got killed by slippage on the Canada universe. Adding AvgDailytot(60)>200000, fixed it with a 2% drop on perf on the US universes (96%->94%). I felt the mktcap filter was enough but can’t take the chance of P123 counting 1,5% or 5% slippage on every trade !
Customs universes may have been restricted after I created mine. I was there early, before Quant crashed the server [:p]
Getting an error when running sims currently:
“Universe ALLFUNDCAD has 0 stocks on 12/30/2006”
Canada is only on the BETA server. See first post in thread.
This is really awesome. Biggest risk to the future success of quantitative investment strategy is inadvertently building the strategy in a way that takes advantage of trends and coincidences that will not obtain going forward. To reduce this risk, it’s really important for us to have ways to test our strategies out of sample. The Canadian data set is a great start. Thanks Marco!