For intraday pairs I think its better to look at standard deviations of intraday moves like open-high, open-low, high-low, ATR meassured over the last 20 days for example. Bollinger Bands suck intraday because of the gaps you see many times in pairs at the open and this causes the moving average to be out of line.
Thanks for the reply. Could you elaborate on the open-high, open-low, high-low system? To summarize what I found from replies and PMs (and again I really appreciated everyone's effort), BBand doesn't work well. From my backtests in the morning BBand is too abnormal due to gaps, and in the afternoon the band is too narrow to take any trades. So there are two hypothesis im going to test 1. calculate A*ratio - B (thx Rehoboth), while ratio = SPY or QQQ's gain/loss from yesterday's close. Then get in a trade when the spread acrosses BBand, and get out when it reverts to the mean. 2. calculate A1/A0 - B1/B0 (thx Bob111), while A0 and B0's prev close price for the stocks, and A1 B1 are current prices. Then get in a trade when A1/A0 - B1/B0 > x%, get out when A1/A0 = B1/B0. Then try different values of x% for each pair to find an optimal one. This method eliminates the conventional BBand altogether. I am an amateur coder so its gonna take a while, I will paste the results back tomorrow.
besides the two variations mentioned above, i still really want to try the high-low open-high thing. Both Rehoboth and Mercury77 mentioned about it but I am not 100% sure as where and when to genetrate entry and exit signals. Could someone elaborate on that please? Thanks.
Let say the average range (high- low) for a pair for the last 20 days is 0.80 cents, you can calculate averages and the SD in excel. The same goes for the average Open-Low (for the pair) and High-Open for the last 20 days and the SD's for both. This gives you some numbers to work with and come up with a strategy. You have to think out of the box with these pairs. You can trade break outs too ( buy the strong and sell the weak) instead of mean reversion.
Thats not what I meant/said. The ratio is predetermined for the day based on previous data. Also doing it verse the spy or qqqq is two general. For example KIE for insurance companies is better at creating a ratio rather then the SPY. Basically the ratioed spread is a prediction point of the were the spread should be based on the movement of the index(kie). So if the ratio for stock a and stock b is 1, then if KIE is up 2% from a point in time(close to now, open to now, low to now etc...), then both stocks should be up the same amount, but not necessary 2%. Also you need to account for how quickly the skew between 2 stocks happens. Do you want to mess with a drifter pair or a pair that made a hard move.
Thanks mercury I came up with a new sheet yesterday to test the aforementioned formula of percentage spread where I long the ratio when A1/A0 - B1/B0 < -x% and short the ratio when A1/A0 - B1/B0 > x% and get out when A1/A0 - B1/B0 = 0 where A0 and B0 are reference points I tried to use prev close and today's opening prices for both A0 B0. Most of the pairs (that worked for many others, like ABX:GG, FDX:UPS, CTSH:CTXS, basically all the well known pairs) didn't work. Only 2-3 pairs show a little bit of profits over the last 3 months. I honestly don't know what's wrong with the system, seems to work for others. Anyways I will play around with the high-low, high-open, and open-low thing and see if there's opportunity in that. What I can do is to get a mean and standard deviation of the spread as you said, and plot a average line and band around it. The only question is do I use a moving average when calculating high-low range or do I take a historical average and apply it for every day? If a moving average is used everyday there will be a gap and the band will tend to draw a big circle on the graph, if I use a fixed average then the band is two horizontal lines.
It still seems like you are not consider time/movement of the pair. You cant throw a blanket strat across every pair out there. Those days went the way of quoting in 1/16's. If pair, with whatever ratio you are using, moves x amount in y amount of time, baring no news, there is % of what it will revert. As Y increase the chance that you have a drifter increases. Ratio, X and Y need to be defined before the day starts. As mercury said it doesnt always have to be mean reversion.
http://www-stat.wharton.upenn.edu/~steele/Courses/434/434Context/PairsTrading/PairsTradingQFin05.pdf http://www.fma.org/Reno/Papers/LOPpairstrading_FMA.pdf http://www.f.waseda.jp/yotsuzuka/Courses/Pairs Trading.pdf Only read the last one but the works should give some valuable insight and the biblio for each is worth exploring. Good luck.