On the face of it, this is a simple issue. Is it better to be long overnight or long during the day? However, once we drill down into it, we find the issue to be much more complex.
For now, we’ll look at the simple part of it.
In 2010, has it been better to be long during the day, from open to close, or has it been better to buy at the close to be long overnight, and sell the next open? No commissions or slippage were included in the tests.
Results of Being Long During the Day:
SPY Net % Gain = 1.52%; Avg. Trade = 0.01%; Win % = 55.98
QQQQ Net % Gain = 1.23%; Avg. Trade = 0.01%; Win % = 53.80
IWM Net % Gain = 0.43%; Avg. Trade = 0.01%; Win % = 51.09
SPY Equity Curve:
Results of Being Long Overnight:
SPY Net % Gain = 7.16%; Avg. Trade = 0.08%; Win % = 55.43
QQQQ Net % Gain = 7.94%; Avg. Trade = 0.09%; Win % = 56.52
IWM Net % Gain = 8.93%; Avg. Trade = 0.10%; Win % = 57.71
SPY Equity Curve:
With the SPY up 2.18% YTD, we can see that there has been a significant edge to being long overnight.
I find myself wondering, “What are the implications of this edge?” particularly so because PDS has consistently performed better selling at the close rather than waiting for the next open, which is counter to the results demonstrated by these tests.
And that is where the issue gets more complex. How do we use this edge when trading individual stocks? Do we create lists of stocks that are highly correlated to the indices and use this research as a guide to buying and selling? What type of environment would merit being long in the daytime, and when would we want to switch to being long overnight?
Hi Wood,
Great piece of information. I haven’t been on the site for a little bit because I had to get a new computer:) Anyway, a while back you mentioned that you use pattern recognition software by I believe Ami brokers. Do you still use the pattern recognition and do you tie it in to your power dip system?
Hi Jacob,
I do use AmiBroker but it isn’t really pattern recognition software. You can create code to recognize patterns. I do use some code to recognize cup with handle patterns.
The power dip is not pattern recognition, in the most widely used sense of pattern recognition, but I guess in a loose sense, it might be. The power dip simply finds stocks in an uptrend, that have pulled back. The charts typically look different from one another, but in the end, it is still a pattern (loosely?) Hope this helps!
http://thepatternsite.com/visualcpindex.html
The complete list of chart patterns.
hi Wood,
regarding your observations in your post, i have observed something like the same in one of my strategies where i sell into strength (of a day or two). However, in another where (loosely) it buy based on a temporary setback (mean reversion play) there seems to be a positive edge selling into open.
I have rationalized it (based on two observations i consider it a rationalization, not a scientific experiment) to say that the market overnight is rewarding mean reversion, and is also mean reverting.
That may explain your observation on power-dip (provided you are selling into a couple of days of strength there as well ) – or it may not.
However, i am still selling into close, mostly since i feel the executions are better there, but thats a different story.
I have not spent any time quantifying this observation further due to having other priorities right now., but Just thought i would pass on my 2c.
best wishes,
bgpl
In my opinion if a pattern is identified but the cause of the pattern can’t be measured then all that has been uncovered is a coincidence.
Why not try testing this over the last 50 years on several global markets and see if you can identify a way to identify periods when it is likely to ring true. If the pattern is real then its strength must be measurable otherwise it can come and go without warning and I don’t think it can be valid for use in a trading system.
Cheers
Derry
Derry….don’t you think if you take the average of 50 years, you’ll just get something in the middle which wouldn’t be useful (much like Sept is historically the worst month of the year, but this Sept has been a huge winner). If anything history should be broken into uptrends & downtrends…or Nov-Apr vs May-Oct…or some other way to make results more useful. JMO
Grrr, it was like you were reading my mind.
I have not tested “the overnight edge yet”, but the intra-day edge looks pretty solid when you add in a filter for stocks with higher unadjusted closes (versus stocks with lower unadjusted prices). I am still trying to account for how volume should be filter for a more realistic expectation.