Overnight RSI(2) Strategy
I’m trying to get a better understanding of trading opportunities in the overnight market (i.e. after the close and before the open).
Below is a simple SPY strategy meant to illustrate the uniqueness of the overnight session (but for reasons I’ll talk about afterwards, this one is a nonstarter).
[growth of $1, logarithmically-scaled]
In the graph above I’ve assumed that each day just prior to the close we calculated RSI(2) for SPY (RSI(2) is an oft-used short-term overbought/oversold indicator).
If RSI(2) closed below 20 (oversold), we took a 200% (leveraged) long position in SPY at the close. Between 20 and 80, a 100% long position. And above 80 (overbought), we remained in cash.
One small quirk: if the market was in a downtrend, as defined by the 50-day moving average closing below the 200-day, we halved the position.
This is an overnight-only strategy, so all positions were sold at the open. I’m being (unrealistically) generous and assuming no transaction costs, slippage, or margin costs.
Numbers for the number lovers…
There’s clearly something special about the overnight session.
As we’ve talked about ad infinitum on the blog, swing trading with RSI(2) has been a killer strategy for most of the market’s history, but performance has been middling for the last few years as that low hanging fruit was traded out of the market.
But as we’ve also talked about, the daytime and overnight sessions are two entirely different animals and it would seem that RSI(2) still plays at night.
The problem with this particularly strategy is that, because the overnight market isn’t particularly volatile, there isn’t much meat in the daily returns. Returns may be consistent, but they’re too small to reliably overcome trading frictions.
In the simple test above, average daily returns (when invested) were 0.0838%, or $8.38 for every $10,000 traded. Trading frictions and margin costs would easily eat half of that.
So this strategy is a nonstarter, but I think it illustrates nicely that traditional indicators respond uniquely when used to trade the overnight session.
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: Trading Strategies | 25 Comments
I would imagine you have already looked at this or soon plan to, but what about tightening the criteria to presumably pick fewer, better opportunities and reduce costs?
Hello hhohw – good question.
For most of the period tested, a deeper RSI(2) (say, only taking RSI < 20 trades) has been a stronger bullish overnight signal. So yes, more meat in the returns there to overcome trading frictions.
The problem is that that observation has waned over recent years (along with the waning of RSI(2) as it's normally interpreted), so now deep RSI(2) readings aren't much more predictive in the overnight than less-deep readings.
The strategy presented here has continued to perform b/c it's taking advantage of 2 factors: (1) the predictiveness of RSI(2), AND (2) the fact that the overnight session tends to be more bullish than the daytime session, period (regardless of how OB/OS the market is).
So getting back to your original question, I don't think any variation of this simple strategy really makes sense (after accounting for frictions) in today's market.
michael
There is a natural upward drift on the overnight (even in bear markets).
Hello Dave – I’ve talked about that many times on the blog, most recently:
http://marketsci.wordpress.com/2012/05/08/the-markets-are-nocturnal-but-not-this-year/
And as I mention in the comment above, that’s part of the reason this strategy “works”.
michael
If this strategy proves to be truly robust, then one could add some meat by trading ES E-mini futures. What do you think of such a strategy?
Hello Alex – I wouldn’t be comfortable trading the strategy as I presented it here with big leverage, but with intra-session stops to protect against large gaps down, it might make sense.
Side note: the strategy hasn’t been as robust over the last few years as it has been historically (in terms of returns relative to volatility). I think that’s related to this post: http://marketsci.wordpress.com/2012/09/02/disappearing-daytime-vs-overnight-volatility-and-why-that-matters/ I would want to continue watching and seeing how that issue shook out before I committed capital.
michael
Use a Mutual Fund that allow morning sales. I know Guggenheim runs one, and if you’re on a platform that allows you to trade at NTF, then you’ll have zero trade costs. Just a thought.
Hello Sam – interesting thought – I’m not familiar with the Guggenheim funds – I know Rydex also offers an AM trade. Will have to run some numbers on both. Thanks for the smart comment. michael
not really a great suggestion. obviously, trading costs are passed along to customer. you’ll see them reflected in daily nav’s.
Hello Steve – I think the hope was (or at least my hope was) that economies of scale would drive that cost down.
In any case, I ran the numbers and it’s no good. The Guggenheim AM pricing is pretty late in the day (10:30?) and the concept presented in this post no longer even remotely works.
michael
Ha, Rydex became Guggenheim last year – they’re one in the same. Shows you how long it’s been since I’ve traded these. In any case, great suggestion. michael
If one has an account with Vanguard, it’s possible to trade VTI with no commission and with very low fees.
Hello Alex – that would certainly help. Still worried about the declining performance (relative to vol) that I talk about in the “side note” of this comment: http://marketsci.wordpress.com/2012/09/03/overnight-rsi2-strategy/#comment-6911 Would want to take a wait and see. michael
My name is Mike and I want to come back to this strategy. Last days I was playing with it a little bit. First conclusion is/was, I have to say I see more potential then u do. If you are implementing more specific leveraging rules. If you are following the RSI you have the expectation, that if lower the RSI in an uptrend market, then higher the chance for a rebound. So why not betting much more with a RSI at 10 compared to an RSI of 50 ?
I just made a short test without any optimization with following leverage rule :
- MA50 Don’t trade
- Base_trade_size = 100
- if RSI(2) >= 50 and 10 and < 50 then buy 2 x Base_trade_size = 200
- if RSI(2) <= 10 then buy 4 x (2 x Base_trade_size) = 800
If I have the confidence that in an uptrend a weaker RSI(2) will support the market, I should bet on it. My test gave me half of your max DD with more then doubled return. I will make more tests with specific rules for the MA's (trading between the MA's or trading above both MA's etc etc).
Regards.
Mike
Sry .. Something went wrong with my post.
- if MA50 = 50 and < 90 then buy 2 x Base_trade_size = 200
- if RSI(2) <= 10 then buy 4 x (2 x Base_trade_size) = 800
I'm sorry again.
Mike
Hello Mike – I encourage readers to draw their own conclusions about the simple strategies I present here on the blog, and I’m glad to see you’ve done your own analysis. Just a couple of thoughts about your results:
1. In more recent history, deeper RSI(2) readings (i.e. closer to 0/100) haven’t been much more predictive than shallower readings. This is a change from most of the last decade when RSI(2) tended to act like a rubberband (the farther it stretched, the more powerfully it snapped back). I would take a second look at whether it makes sense in today’s market to ratchet up exposure as RSI(2) approaches 0/100.
2. I would be wary of employing significant leverage in the overnight market unless I were trading futures w/ protective stops.
Just my $0.02. Thanks for the comments. michael
I tested the 3:30 trade 3 years ago- on a different basis (i.e. if at 3:30, ES > Open, Buy. Sell at Open next day. This Buy at 3:30 case was better than Buy on Close. However, I recall it was not a good trade in 2010. I will look at again.
Bruce
1. This does work with futures where the transaction costs are much lower.
2. Even better- enter at 3:30pmEDT.
Hello Bruce – thanks for the comment – have you run the numbers on the 3:30 entry? Did it significantly impact results? I didn’t run intraday data, but sounds like maybe I should have…michael
Interesting, Bruce.
Do you have any data (i.e., backtests) using futures for this strategy and entering half and hour before close?
Bruce & Alex – just coming back full circle on this – ran the number of other entry times (15:30, 15:45, etc). Assuming I didn’t snafu my numbers, results look significantly worse than entering at 16:00. Bruce – am I missing something?