to dream
People have long imagined ways to make money while they slept. Happily, it’s not a pursuit I’m particularly bothered by, but as I develop trading strategies, I do make note of different market behaviors that correlate to the time of day. Or night.
In particular, I’ve been looking at various market-breadth ETFs recently as possible fodder for the little dynamic hedger I’ve described before, and I’ve noticed an interesting behavior among several of them…
Like the majority of traders, they do better when they’re not trading!
That is, they actually display better performance at night than they do during the regular trading day; there’s more profit to be had in their gaps between sessions than there is during trading sessions. Below I quantify this observation more thoroughly…
I looked at the performance of a variety of ETFs across three super-simple “strategies”. The first of them trades overnight – buying the close and selling the following morning’s open. The next buys at the open and sells at the close. And the final one is, essentially, buy and hold though it’s implemented as a buy at close, sell and buy at following close. No costs or slippage are included. The table below details the results for the SPY (S&P 500), IWM (Russell 2000) and QQQQ (Nasdaq 100).
In all cases, the ETFs perform better overnight – they make more and have almost precisely half the risk, yielding very substantially better risk-adjusted performance measures (here sharpe is calculated with a constant 5% risk-free rate). These cases are highlighted in yellow in the table. Likewise, the worst performer in all cases was the “day-trader” who bought on the open and sold on the close. One thing to note about the below summary is that all data is taken from inception for each of the ETFs, so their duration varies! (For the SPY, since 1/29/1993; for the IWM, since 5/26/200 and for the QQQQ since 3/10/1999).
I tried the same analysis across a few other ETFs to see if the effect was a global one or a US-specific phenomenon. The results are mixed but likely negative for Europe (EWG), Japan (EWJ) and broad foreign markets (EFA). I’m not going to summarize that data here, but you can grab the excel workbook I used for this exercise here, though I warn you in advance that it’s about 2.3M. (If you find errors – please point them out!)
[UPDATE: .XLS 97-2003 version available here is ~3M]
For the US markets summarized in the table, the conclusion is pretty explicit – the great majority of the market’s profits are attained overnight. Even worse: in spite of the positive returns across all of these markets over the relevant periods, all of them show significantly negative intraday performance.
For long-only day-traders, this is the kind of result that should cause nightmares!
Hi,
I have been reading your blog and even though I am not part of the financial market world at all, it has been very interesting!
Trying to recompute your figures I came up with a slightly lower return for the SPY (with 10.64% annualized return at the end of 17/07/2008) while near exactly the same volatility (which I computed using the log return method).
I did it on the bel20 index (belgian index) and came up with the same results with a 10.79% annualized return and a 8.94% volatility
I am just wondering what’s the impact of the fees on the return do you have a rule of thumb?
Another question: which fixed income rate did you use to compute the sharpe? I used the 6 months CD rate but I don’t now whether it is too cautious or not?
And last but not least, would it be possible to have an excel 2000 compatible version of the file?
Thanks,
Didier
Hi Didier,
Thanks for your kindly comments. Sorry for the delay in responding, but I’ve been sidetracked by a happy family event – the birth of our first child!
I’ve updated the post to include an excel97-2003 copy of the workbook – hopefully it makes sense to you and isn’t too ridden with errors.
I consider this a “study” rather than a strategy – I’ve calculated no fees, comm.s or slippage in this sample. Within StratBox I provide a variety of ways to deal with these factors and I touched on the issue a bit in my May 6th post (”quantifying friction”), but will revisit it again in the coming weeks as it’s an interesting and important part of the puzzle. I also should be able to quantify what I’ve actually been slipping on some of my production strats for illustrative purposes.
For the risk-free rate, I used 5% in this workbook, but that value can be adjusted. For studies like this, I really don’t think it matters much and will use a random value from 0-5% depending on mood, position of planets etc. For a bank it makes sense to use the appropriate interbank rate (eg, libor) and for a retail investor it probably makes sense to use the rate your broker pays you on deposits, a cd rate or something similar. For its use in characterizing a risk-adjusted measure of returns, I don’t think it really matters at all.
Best regards
Hi tito,
I am glad to hear such a good news! It seems like your 9 months investment has grown really well ; )
Thanks for the sheet I see why we have got slight differences now.
For the annualized return, I was using a rolling geometric mean which is slightly different to the approach you are using.
Anyway thinking about the slippage I was wondering whether in this particular case, it was not reduced to a minimum since we would not put any limit to the buy order at the closing and a not limit sell at the opening (in its most simplistic approach)
About the risk free I am not entirely convinced because when one has to compute the sharpe ratio over a certain time period, say your annualized return in 200X was 10% but if for that year, the risk free rate was say 8% it is not the same as a 10% when the risk free is 2%. Of course in our countries with low inflation rates (for the last decade at least) this is not much of a problem but could well be in the future or if you have to transpose your sharpe ratio metric for other countries.
Just my 2cents thoughts
Many Thanks,
Didier
You have no idea what a return – the boy was nearly 4.5 kilos at birth!
I certainly agree with you re: the risk-free rate used in sharpe when comparing different strategies. But if one is using sharpe to compare different iterations of a strategy as it is being developed, it surely doesn’t matter which value is used.