Overnight vs. Intraday ETF Returns

I haven’t done much “googling” before posting, so this topic might have been covered elsewhere but I think it’s  really worth sharing or repeating anyway.

A lot has been written about the source of  ETF returns (some insights might be found here). In a nutshell some analysis found that the bulk of the return is made overnight (return between close price at t and open price at t+1). This is only partially true as it hides some major differences across asset classes and regions. The table below displays the sum of daily returns (close to close) , intraday returns (open to close) and overnight returns (close to open) for most liquid ETF over a period going from today back to January 1st 2000 when data is available. The inception date of the ETF is used when no data is available prior to January 1st 2000.

ETF Daily Rtn Intraday Rtn Overnight Rtn
SPY 53.7% -8.1% 59.2%
QQQ 10.7% -84.3% 93.3%
IWN 81.8% 30.4% 52.1%
EEM 51.5% -42.5% 83.8%
EFA 13.2% 73.3% -61.5%
EWG 77.7% 143.1% -62.6%
EWU 41.2% 132.3% -84.5%
EWL 109.4% 229.9% -110.3%
EWJ 10.4% 115% -107.9%
FXI 72.8% 13.8% 45.3%
EWS 89.7% -83.9% 175.9%
GLD 120.9% 18.7% 101.1%
GDX 29% -270.2% 293.5%
SLV -2.8% -36.6% 39.1%
USO -21.6% 56.7% -79.5%
SHY 4% 10.7% -6.5%
IEF 23.5% 37.4% -13.4%
TLT 37.1% 50.6% -13.5%
LQD 16.7% -36.3% 54.3%

A few obvious features clearly appear

  • For US Equity markets (SPY, QQQ, IWM), Emerging Equity Markets (EEM), Metals (GLD,GDX,SLV) and Investment Grades (LQD) the bulk of the return is definitely made overnight. Intraday returns tend to deteriorate the overall performance (intraday return < 0)
  • The exact opposite is true for European Equity Markets (EFA,EWG,EWU,EWL), US Bonds (SHY,IEF,TLT) and Oil (USO). Overnight returns are detracting significantly from the overall performance.

I didn’t manage to come up with a decent explanation about why this is happening but I’m keen on learning if someone is willing to share! I’m not too sure at this stage how this information can be used but it has to be taken into account somehow.

Below is the code for generating the analysis above.

## thertrader@gmail.com - Jan 2014

symbolList <- c("SPY","QQQ","IWN","EEM","EFA","EWG","EWU","EWL","EWJ","FXI","EWS","GLD","GDX","SLV","USO","SHY","IEF","TLT","LQD")

results <- NULL

for (ii in symbolList){
  data <- getSymbols(Symbols = ii, 
                     src = "yahoo", 
                     from = "2000-01-01", 
                     auto.assign = FALSE)

  colnames(data) <- c("open","high","low","close","volume","adj.")

  dailyRtn <- (as.numeric(data[2:nrow(data),"close"])/as.numeric(data[1:(nrow(data)-1),"close"])) - 1
  intradayRtn <- (as.numeric(data[,"close"])/as.numeric(data[,"open"]))-1
  overnightRtn <- (as.numeric(data[2:nrow(data),"open"])/as.numeric(data[1:(nrow(data)-1),"close"])) - 1

  results <- rbind(results,cbind(
    paste(round(100 * sum(dailyRtn,na.rm=TRUE),1),"%",sep=""),
    paste(round(100 * sum(intradayRtn,na.rm=TRUE),1),"%",sep=""),
    paste(round(100 * sum(overnightRtn,na.rm=TRUE),1),"%",sep="")))
colnames(results) <- c("dailyRtn","intradayRtn","overnightRtn")
rownames(results) <- symbolList

As usual any comments welcome


  1. Arun S says:

    Interesting finding. Wonder if this is a statistical anomaly – seems unlikely given the geographic pattern. Some type of sensitivity analysis, maybe different time subsets, to check the findings will be easy to do and strengthen the findings.

    If the pattern does persist, wonder if the underlying assets of these EFTs also behave the same way.

    My thoughts. And thanks for an interesting post.


  2. Nice post. You should check out the rolling correlation structure of intraday return to overnight return.


  3. etf_trade says:

    Limited resarch out there: http://www.palgrave-journals.com/jam/journal/v12/n2/abs/jam20112a.html might fit. They can show that this effect is significant but they are not able to explain it: “Institutional investors who conduct all
    trading at closing prices may want to seek to
    rewrite their contracts to allow for
    liquidations at open prices Other investors
    who tiade on a frequent basis, such as hedgers
    of derivative portfolios. may want to time
    their trades for better pmfitabilit Finally, if
    semiprokssional investors are liquidating
    their undiversthcd portfolios at the end of the
    day and are causing this effect, the question of
    why market participants do not take
    advantage of this behavior remains open.”, p. 143

  4. martin Bauer says:

    Using ETFs in different time zones

    I was looking at EWG – MSCI Germany
    Obviously between the DAX and EWG there is a high correllation. Nevertheless the EWG is the USD version of the DAX. using ^GDAXI I get a very diferent view
    77.1% daily
    2.3% intra
    70.2% overnight

    I reckon that is worth to mention it

    happy trading

  5. russ says:

    This guys seems to have looked into this, although I suspect the REALLY interesting stuff has to be paid for…


  6. hoojammyflip says:

    where do the closing prices come from…lots of manipulation in the markets being written about, banging the close

    what happens if you subset the data differently, so instead of taking the price at the close, you use the price from half an hour before, when the market is actually liquid?

    • The R Trader says:

      Closing prices are from Yahoo (look at the R code). Not the most reliable source I agree but not too bad either escpecially for the most liquid ETFs. I traded those ETFs myself with institutional money and you can really get the closing prices (most of the time). So I’m confident that the pattern observed is real. Regarding your suggestion: It would be nice to see the results with data 30 min before the close but I don’t have the time to collect the data and reproduce the analysis. If you feel eager enough, please feel free to do so, I’m keen on learning.

  7. Roddy says:

    I have been thinking about this and whether it is really exploitable:
    (1) I wonder if the overnight performance is the reflection of e.g. dividends being paid and then the opening price being marked down accordingly – not sure but will be worth doing the backtest on both the close and the adjusted close (and adjusted open) columns. I think there is an R function adjustOHLC which would help in this. It is also worth considering what the tax impact on the dividends would be and how that would impact the returns.

    (2) A typically large institutional investor will pay 10 bps for trading. I think this would eliminate much of the trading advantage.

    (3) I suspect that the bid – offer spread; and slippage would eliminate much of the rest.

    However point (3) is speculation on my part and would love to be proven wrong.

    • The R Trader says:

      Thank you for reaching out.
      Tax, Transaction costs & slippage are valid points as they have not been taken into account in my post. But given the size of the differences between Intraday and Overnight returns I suspect there will still be some alpha left after all this. This is especially true for the US equity market. One factor that is not in my post is whether to consider the aggregate ETFs prices or just the primary exchange (I downloaded data from Yahoo and I don’t know which figure they use). The real test would be to conduct this analysis with the primary exchange only in addition to what you mention. it might yield some more differences as well.
      My main concern isn’t actually so much on trading costs but more on why this is happening. I still didn’t come up with a decent explanation and I don’t want to use something I don’t feel comfortable with. But this is just me…

      Hope this helps

      • Roddy says:

        I think one needs to consider how trading and information flows during the day. Most companies pre-announce any major information before the open – so if anyone is going to react they often place orders to buy on the open. However during the day – probably sell orders occur causing stocks to drift from the open price. Thus overnight trading really means ‘front-running’ (in the loosest sense of the meaning) any pre-open positive announcements. I guess another way to consider this is the skew (positive vs negative) of pre-open results or other announcements.

        What do you think?

        • The R Trader says:

          Hi Roddy,

          What you describe makes sense for the US market but it doesn’t explain why the pattern is the exact opposite for Europe.
          I see it more as a cross-regional pattern than a stock specific pattern. The European market being opened well before the US it absorbs all the news before the open in the US. As markets tend to go up over the long run the European market goes up intraday (on average). Once the US opens, it catches up with Europe at the open, hence the overnight returns being far superior to intraday returns. Conducting the same experiment on futures rather than ETFs would probably be very informative.

          Hope this helps

          • Roddy says:

            The point about different patterns in Europe vs US is very interesting. I am looking at this issue again and will consider this. Will also be worth looking at how Asia absorbs information. I have an hypothesis that different sectors might have different behaviour (or different sized companies).

  8. The Blind Daytrader says:

    I think that has to be it. The fact that the future keeps moving after the equity close, and that after the first tick on the open the ETF usually has to catch-up.
    The world doesn’t sleep while the U.S. market for ETFs is closed. So there must be an immediate digestment of overnight news, overseas actions, the future’s movement, whatever.

    Seems pretty reasonable, actually–the market makers have been well aware of this for a long time. Watch what the ES does before the open, and you’ll know what SPY will do after it. At least, so I’ve heard. I don’t have a source for instant to instant futures pricing to verify.

  9. bitfool says:

    Thanks for posting your R code, I’m just getting started with R and portfolio modeling. I ran your code just now since Jan 1 2015 instead of for many years, and the results are much less pronounced. So that got me curious, and I ran some quarterly tests, and for 2013 SPY and QQQ (just to look at the first two) are totally different from the long-term results you posted… and are not even fully consistent Q to Q. For SPY, daily always wins, but intra vs overnight are wildly different. More consistent is QQQ, daily always wins, and daily is roughly twice the return of either intraday or overnight returns. I suspect there’s no easy lessons here at all, but the surprising resiliency of some findings bears attention.

    2013 Q1
    SPY 6.7% 6.0% 1.4%
    QQQ 2.1% 1.3% 1.5%

    2013 Q2
    SPY 3.6% 0.9% 2.4%
    QQQ 4.8% 2.1% 1.9%

    2013 Q3
    SPY 4.9% 1.3% 3.7%
    QQQ 10.7% 4.7% 5.9%

    2013 Q4
    SPY 8.8% 7.3% 2.3%
    QQQ 10.1% 5.6% 5.5%

    • The R Trader says:

      Hi Paul,

      Thank you for reaching out. I’m not really surprised these things are not really consistent overtime but the pattern is there over the long run. It’s like any quant strategy: you know the pattern is there but you don’t really know when it will materialize hence the need of being disciplined in the approach. Once you defined a strategy you’re comfortable with stick to it. In this particular instance however this is not a strategy and I still don’t know how this can be exploited.


      The R Trader

  10. nice info The R Trader about sahre market intraday trading software

Leave a Reply