That historical SH>2 pair begins to degenerate after Mar. Though that pattern could be recognized. Selected indicators on ratio chart, training period selection, change in the OU process paramter, half life estimation, regime switch model could be the solutions to computerized filtering those pairs. Even by naked eyes, I would say their movement pattern are quite different before and after Mar. If you are lucky enought not getting the entry signal just before the pair begin 'the ugly trended the other way' and 'choppy up-down around +-2 standard' movement, everything is fine.
Thanks Cooper, Apologies for the staccato nature of this post, I had written a longer version in an edit of my last post but, infuriatingly, I lost it in the 30mins time-out rule when I tried to post. Agree 100% on the need for discretion in selecting pairs. Pairs are pairs because their strategic end markets are similar and therefore there profit drivers will be NOT because their price charts happen to look the same. I wouldn't short the next move in coast line of Miami just because it's shape diverges from the chart of IBM! I'm basing my trading on discretionary & quantitative filters to select the universe of stocks and then developing a quantitative entry/exit point system in order to trade. It logically follows that I should have a quantitaive way of defining when the realationship breaks down and the pair is no longer a pair. This problem has been discussed on this thread, but as yet, I have not seen any articulation of an adequate solution to it. At least to me! To get some discussion going, I will tell you what I have tried. I have backtested my universe of pairs, using a relatively conservative entry/exit signal system and used the following criteria. 1/ 2% stop loss 2/ 3% stop loss 3/ no stop loss. Note the '% stop loss' just refers to the percentage loss on the trade. I appreciate that the individual volatility of the trade should be taken into account, but I am looking for a simple system here with a consistent trade size. My results are linear in their optimisation... 1/ 2% stop loss. W/L ratio 1.16 W/L prob .67 2/ 3% stop loss. W/L ratio 1.39 W/L prob=.7 3/ No stop loss. W/L ratio 1.66 W/L prob=.72 based on 100+ pairs measured over 2 years. Clearly the stop-loss tends to erode the potential return. Also, all three expectations are strongly positive. I have real time tested the conclusion via trading (with the optimal no stop loss option). Admittedly, my sample size is still small (35 trades so far) but already, I am finding that I am getting too many larger losing trades. This maybe due to... a/ small sample size error b/ my universe of pairs are 'pre-selected' and maybe biased and therefore not indicative of future results. My suspicion is that 'b' is the reason. Furthermore, applying the same logic means that even the 3% option may not work along with the theoretical results. I am happy to be proved wrong! For other examples of relationships that break down (like UPS FDX) see CNA ALL & LOW HD, or even my current bete noire of AMAT-LRCX & DCI-PH. Applying discretion to exiting trades is, for me, fraught with difficulty because when we go through bad patches (as we all will) it will be very difficult to retain focus on the discipline of the 'trading compass'. Am happy to share this info, and learn more about how others deal with this problem. It strikes me as being THE integral issue behind pair trading.
Another reason for your underperformance in live trading could be that with the furious rally that has been underway, stat-arb has underperformed lately (esp. in July and September, so far.) AS the market rises you need more and more factors to explain the variance of stock returns which decreases the profitability of stat arb strategies.
My view has always been that returns in Pairs will crush a bear & sideward market....But struggle to keep up in a raging bull market... This is only the COMPARISON of returns, not the profitability of the system itself. If you say that stat arb struggles in a rising market, then the same principals would apply in a bear market. Or any trend for that matter, which I don't believe is entirely correct. That's the crux of it though. Returns are supposed to be completely independent of market activity. It doesn't take a brain surgeon to work out that returns won't be as attractive when comparing to a market that has put on 50%....
Definatley, stops are the hardest part. I would love to hear others thoughts. My view is you simply need to keep in tune with the market, and get a feel for price action and develop a fundamental edge. In ordinary circumstances we can just rely on the stats and numbers that our calculations tell us...85% of trades fit in this category. When things break down is the toughest part of pair trading Recent example: I have been in CVH / UNM since last Wednesday. It got pole-axed on Friday and I was down about 7% on the pair. A class action had been launched against CVH but no real specifics & stock was down 6%. I've seen a few knee jerk re-actions like this before but with nothing specific on the case, it seemed a little overdone. Doubled my position in the pair... Last night CVH 4%+ UNM -1.5% the pair is now in profit after staring down the barrel of a hefty loss... I WISH I COULD REMOVE ALL DISCRETION AS THIS WILL BACKFIRE FROM TIME TO TIME.......AND IT IS A TOUGH GAME TO PLAY It leaves soooo much room for internal bias But I don't think anyone has an answer...
the difference between bull and bear markets is that bear markets have higher volatility which helps statarb, and the lower volatility in bull markets makes it underperform (relatively speaking). Look at the low volatility bull market of 2003-2006 for example. So, there isn't perfect symmetry there.
cooper1308 ej420, Good points there. I think you might be right about stat arb underperforming in an extended bull market, and I think more real-time and back-testing is needed here. However, the question for me is whether there will still be positive expectation in this scenario. It is not a relative consideration. The point being that by using leverage, you are still likely to achieve greater risk-adjusted returns than being long-only, either levered or not. Even in a bull market the mini corections can significantly erode capital if you are levered. Avoiding market risk is an end in itself.
I agree higher volatility should provide better conditions to trade......VIX is lower but still at 2008 levels....This aside, the market has still gained 50% in 6 months and that alone should provide enough movement.. I don't agree that the fact it has been a "rally" that performance has eroded.... Would you agree a market that drops 50% in 6 months would provide roughly the same opportunities as a market that rallies 50% in 6 months. Yes generally markets don't behave this way ("Up the stairs, down the firemans pole")......But we are experiencing a market seen once or twice in the last 20 years
Anyone knows of a broker that enables a one-click limit order for a pre-determined pair of stocks (without an API)?