Thursday, September 25, 2008

Day and Night Sessions: This Bear Isn't Nocturnal

I sat down with a young day trader before the market open and asked his strategy for the morning session. "I'm bearish," he explained. "The markets have all been down overnight."

"Do you know what the correlation is between the overnight change in the market from close to open and the change in the day session?" I asked innocently. He shook his head, seemingly surprised that someone would actually ask a question pertaining to data. "It's -.08 since 2007," I explained. "What happens overnight is not related to what happens during the day."

Our young trader was unusually naive when it came to his short-term trend following, but the fact remains that many short-term traders find their opinions colored by what happened overnight. In reality, these function as independent markets: what happens overnight is not predictive of what happens during the trading day.

A nice illustration of this independence is captured in the chart above. I created two indexes, both set to a value of 100 at the start of 2007. The first index (blue line) simply adds the SPY points from close to open to a running cumulative total. This is the "Night Market". The second index (pink line) adds the SPY points from open to close to a running total. This is the "Day Market".

Since 2007, the Night Market has gained 9.4 SPY points (the rough equivalent of 94 S&P 500 Index futures points). During that same time, the Day Market has lost 30.23 SPY points (about 302.3 S&P 500 Index futures points). For all practical purposes, the entire bear market during 2007 has occurred during the trading day; not during overnight trade.

But if these markets are truly independent, might we be able to best predict what will happen during the day by limiting historical investigations to previous day sessions? Could we predict what is likely to happen overnight by running studies on recent night sessions? There's plenty of room for original and interesting research by segmenting market days. More to come...


Joshua said...

It seems like the day/night decoupling provided a good advance warning of the market topping. It would be interesting if this happened as well in beginning 2000 and before other bear markets. Probably not enough instances for statistical significance though.

I'll test it out and report back. I've used a lot of your testing of historical edges in designing my own indicators & strategies. Time to give back. I've read your blog for almost 2 years before very recently starting to trade the S&P e-mini. I always soaked up the "entry & exit" type edges while not paying the psychology posts much mind. THAT has changed. I definitely think my learning curve has been and will continue to be much quicker primarily from reading your blog. Thank you for that.

I'm also grateful I'm learning in such volatile times. Baptism by fire, right? :)

Brett Steenbarger, Ph.D. said...

Thanks, Joshua; I agree--if you can survive your learning curve in this environment, you'll be well prepared for dealing with future crises!


Alex O. said...

I've gone back to 1993 to compare the day/night gains in the SPY, and can't yet see how a strategy based on this effect could work. Here's the summary:

1) The nightly gains 93-08 are much higher than the daily gains. If you had started with US$ 10,000, you'd now have 46,800 if you had only held the SPY at night, and only 5,800 when only holding SPY during the day.

2) The 'nightly' SPY had its peak in mid 2001 and its bear ended early 2003. The next peak was in november 2007.

3) The 'daily' SPY had its peak in early 1998, it bottomed in mid 2002, had gains until 2004, stayed flat until mid 07 and then turned down again.

4) Only in the following periods the rolling 200 day returns of the 'day' SPY were higher than from the nightly SPY: 3/95-3/96, 8/01-6/02, 10/02-2/04, 2/07-7/07.

I guess you could say that day trading in the past 10 years has been a risky business, compared to a theoretical "buy and hold" of the nightly SPY.

Joshua said...

Here's what I found so far.

( will add details in a few days, I'm pretty busy - sorry )

This seems to only be a property of SPY. Assuming my data source is reliable, SPY's overnight change is very positive compared to its daily change. But this doesn't happen when you look at the e-mini or the actual S&P index. There doesn't seem to be much of a significant overnight change in the underlying index or e-mini, only in SPY. I'll get more in-depth soon. I wonder if this is a "unsophisticated retail trader" effect or if there is an arbitrage play here.

Oh, and no, I don't think there's any bear market warning. Looking at a daily chart beginning in 1995 SPY's overnight vs. daily change lifts off immediately and doesn't top out until about 2005 or so.


Gabriel said...

In mid-july 2007, the first of the really massive carry-trade unwindings of the yen began. That's when Fannie and Freddie's Fed takeover was announced and the dollar's stability became questionable. That's right at the point that the day/night decoupling occurred.

So those 200 trillion yen were swapped for dollars on a long-term basis less and less. But there was still short-term trading. You pick up the LIBOR spread on dollars and yen (about 4%apr) if your borrowed yen are in the form of dollars overnight. So it makes sense that there would still be interest in carrying dollars overnight and buy US commodities or securities short term, then selling those securities in the day market to unwind the currency trade.

It is certainly interesting.

Joshua said...

Sorry for the delay...

These reults are from Jan. 3rd, 2007 to Sep. 18th, 2008.

S&P cash index
Overnight LOST 23.35
Daytime LOST 192.74

SPY ( 10 shares )
Overnight GAINED 1.30
Daytime LOST 214.30

S&P E-mini
Overnight LOST 44.00
Daytime LOST 185.50

I had to create 390 minute bars for the E-mini data because my software uses the Globex open and close for the e-mini. This might cause a variation in the end result. Regardless, I think the take-away is to be careful using SPY data for analysis when you trade the e-mini.