I know that I promised more about designing adaptive strategies, but I need to take a quick step back because I received some very good feedback I want to address from fellow blogger Damian Roskill of Skill Analytics related to this topic.
If you recall, earlier I discussed a static strategy for trading the US Dollar Index (USDX) which I then converted to a much more effective adaptive strategy. The results were as follows (static in blue, adaptive in red):
[logarithmically scaled]
Damian retested my static strategy and got a very different result (my results in blue, Damian’s in red). Note: all of the following charts are for the initial static strategy, not the later adaptive one.
[logarithmically scaled]
What gives? Prior to the mid-2000’s when my version was doing pretty well, his was bleeding. In fact, in Damian’s test, the reverse of my rules would have made for a pretty good trading system:
[logarithmically scaled]
On first blush, our data is identical (my data in blue, Damian’s in red). Can’t spot the difference? On this graph, for all intents and purposes, there isn’t any.
But there are actually some very subtle discrepancies not captured by the graph. On any given day, the average difference in closing price was about 0.2% and about 12% of days experienced a difference of more than 0.5%.
Why? Damian utilized futures spot prices for the USDX from the NYBOT. Was there anything wrong with that? Absolutely NOT. It’s very feasible that a trader would have executed this strategy using futures.
What’s the point?
The point is that when trading very active strategies that are exploiting very fine observations about the underlying asset, it’s important that we test the strategy on the specific asset being traded because that observation may have already been traded out of that specific asset.
This is akin to taking an active S&P 500 index strategy that might work very well on leveraged funds (because of the very low tracking error) and blindly applying it to the SPY ETF – it often doesn’t work at all.
Big thumbs up to Damian for spotting this. If you aren’t plugged into Damian’s blog, I suggest you do so. He doesn’t post enough (for which I berate him often), but I’ve seen a lot of his system development work and consider him top drawer.
Happy Trading,
ms
To stay up to date with what’s happening at the MarketSci Blog, we recommend subscribing to our RSS Feed or Email Feed.
Filed under: Evolving Markets, Trading Strategies, US Dollar & Currencies | 5 Comments
Hi, Given the comment above “This is akin to taking an active S&P 500 index strategy that might work very well on leveraged funds (because of the very low tracking error) and blindly applying it to the SPY ETF – it often doesn’t work at all”
and that YK says “The strategy is intended for trading leveraged funds such as those from Rydex, ProFunds, or Direxion, but could be applied to almost any vehicle that closely tracks the indices above”
I see a bit of a contradiction? Is it right to say you can trade “any vehicle ..” with the YK signals, or is it essential to use the stated (leveraged) funds?
But also, and not knowing how the signals are meant to be used exactly, the whole idea of “executed at the following close” seems to suggest that, as the definitive close price can’t be achieved whatever the instrument, surely the difference in actual instrument used is a bit 2nd-order?
That said, I’d also like to see the tracking differences if someone has a link to a comparative table?
Please accept my apologies for my first comment on this excellent blog appearing “critical”!
RE to DaveT: no need to apologize – I think that’s a very insightful observation.
I made the comment RE: the YK strategy’s applicability to intraday vehicles based on my own tests. I think we’re able to mitigate a lot of the difference in performance between the index (which the leveraged mutual funds track) and a tradeable vehicle (ETFs or futures) when the trader is still taking the signal based on the index, but then applying it to the ETF/future. I hope that last part made sense. The YK strategy requires a trader to make a decision EOD as to whether the index will close up or down for the day and that decision is still based on the underlying index itself (not the ETF/futures).
Of course, this ignores what I see as a far bigger issue introduced with intraday vehicles – commissions. I have a sneaking feeling that a number of folks are following the YK strategy in far too small of accounts are losing far too much to transaction costs.
Thanks for the thoughts. Again, very good question.
michael
Michael, thanks, I think…
I’m in the UK and have US and UK trading accounts, but I don’t trade “leveraged multual funds” – just a thought, do they only trade on an “EOD” price basis (not like an intraday stock or ETF)?, that might account for some of my ongoing confusion.
Perhaps, because of this and also to examine the commission issue (which I had wondered about but hadn’t got around to yet), you could write an article showing a full “worked example” of applying an example signal on an actual trading day and the different instruments, this could then lead on to a commission compution (introducing number of trades) and show the effect on various account sizes, and compare to using the managed service perhaps? [And with your other hand you can continue to type something creative :-)] That would also help me, especially as I need to also look at the use of the UK-based “spread-betting” alternative (tax-free, leveraged, over-night trading, but with largish spreads / commissions).
Also, stop losses, I didn’t see a mention, do you use them, if so, is there a guide to the maximum risk (for my account size consideration)?
Thanks for your time.
RE to DaveT: all good questions. You’re correct, the leveraged mutual funds only trade EOD (except at Rydex where they trade twice a day). The upside is that they track the underlying index very closely and there are no transaction costs for changing positions (assuming you’re trading directly with the fund company). The downside is that they don’t allow for intraday orders (so no stop losses). The idea of posting a study trying to simulate running YK and/or Scotty over ETFs has been on the to-do list for a while. Hopefully at some point this year I can clear enough off the plate to knock it out.
All the best,
michael