The Equity Overnight Anomaly ETFs
Bad luck, overfitting, and transaction costs
TL;DR
The overnight return anomaly became much less anomalous around 2009
The failed ETFs designed to capture it suffered from horrible timing, but also transaction costs
Transaction costs are much greater than fees, and also greater than fees + (ask-bid)/2
The overnight equity anomaly is that most of the total equity returns are generated from the close to the open [edit: I wrote open-close by mistake before]. When I was an active blogger circa 2011, I posted about the overnight anomaly (noted first by Tom Anichini). At that time, the total return for developed country stocks was generated from close to open, while the risk associated with these overnight returns was less than that of intraday returns. You got risk but no reward intraday. This implied a dominant strategy where one gets most of the equity premium with one-third the risk.
In 2022, AlphaTrAI (now Praxis) introduced two overnight ETFs that tried to capture this anomaly. Their two ETFs focused on the SP500 and Russell 2000: NSPY and NIWM. They lasted a year. The chart below illustrates the behavior of overnight returns from July 2022, when the funds commenced, to August 2023, when the funds were closed. The black line, representing the overnight SPY return, highlights abysmal timing, as the overnight returns were negative while the total returns were positive—an unprecedented occurrence over 12 months. These funds operated in that rectangular period, which is expanded on the right.
However, upon closer examination, this was not simply a case of bad luck. The anomaly had significantly diminished over the past decade, and the pre-2009 data dominated the aggregate data.
The timing of my blog finding (2011) is standard for any quant who spends time looking for these things: I top-ticked it. However, I noted at the time the anomaly would probably not cover transaction costs. It was still a striking anomaly to the standard risk-return theory, in that most risk is generated during the open-close period. Since 2009, it has become more of a curiosity than a striking phenomenon. My best guess is that electronic trading, including sessions after the official NYSE close, made the distinction between overnight and intraday less clear.
A strategy that trades every day generates significant transaction costs, even for trading the most liquid equity asset. If we examine the nightfund’s cumulative underperformance relative to the overnight returns derived from the Russell 2000 and SP 500 ETFs, we see clear evidence of trading costs. Note the linear cumulative underperformance, which indicates a daily drip.
The cumulative underperformance was 7.6% and 1.7% for the Russell 2000 and SPY nighttime ETFs, respectively, given that the strategy traded twice daily—shorting at night and covering at open—resulting in two complete NAV turnovers every day. Over their 269-day existence, this implies transaction costs of 1.4 and 0.33 bps for trading the Russell 2000 and SPY futures (RTY and ES).
The SPY futures contract is the most liquid equity asset in the world. Given that the NSPY was only $3.7 million, the fund manager could easily cross the spread, which is typically $30 million for the best bid/ask size. Fees were around 0.06 bps, and the half-spread only 0.25 bps, which approximates the 0.33 bp trading cost implied by the NSPY underperformance.
For the Russell 2000, things get tricky. Given the 7.6% underperformance, using similar logic, one generates a 1.4 bp transaction cost. The size of the NIWM ETF was a mere $1.7 million, but the best bid/ask size for its futures is around $2 million. So, in theory, it seems like one should not have to pay more than half of the bid-ask spread to make this trade (plus the fee). Its fee was considerably higher, at 0.16 versus 0.06, and the tick average was double that of the SPY, at 1.0 versus 0.5 bps. However, this only yields 0.66 bps (0.16 + 1.0/2), which is less than half of the 1.4 bps trading cost NIWM experienced.
Market makers are fast, and if you try to swoop up virtually the entire top bid/ask, chances are you will lose, especially if you are a mere fund manager as opposed to a specialist high-frequency trader. You can pay a high-frequency shop to access the entire best bid/ask, but that would cost you at least 1 bp, which in this case would be even more expensive. While some might consider this a significant inefficiency, it is worth noting that spreads were typically around 60 basis points for most of the twentieth century. Complaining about paying 1.4 basis points instead of 0.6 basis points seems rather ungracious.
This highlights two important points about transaction costs.
1) Fees are a fraction of total transaction costs. I worked on a high-frequency trading desk, and the fees were always around 10% of transaction costs for liquid stocks. For the SPY futures, the fee was 20% of the total transaction cost, but this is for the most liquid equity asset in the world. For everything else, such as RTY, AAPL, or any other S&P 500 stock, fees as a percentage of total transaction costs will be significantly lower.
2) The bid-ask spread to the mid is a lower bound for price impact or slippage cost. On most stocks, one needs to trade a significant fraction of the best bid/ask size, in which case, you should expect to pay more than that. The market-makers will move away.
The bottom line is that for trading liquid non-ETF equities, which do not have futures, transaction costs are approximately 5 basis points. One should expect to pay at least that to trade spot Bitcoin or Ethereum on Binance, which has the largest crypto/fiat spot markets in the world (~$2B/day). Coinbase, the US’s largest crypto exchange, meanwhile, generates only 20% of that volume. In contrast, the RTY trades $19 billion per day, and AAPL stock trades around $10 billion per day.
A recent paper on CEX-DEX arbitrage assumed the arb could trade out of their crypto position at the mid on Binance, paying only the lowest-tier fee of 1.725 bps. That’s delusional because it implies one can trade at the mid. If everyone could trade at the mid, liquidity providers would make zero profit, which is not an equilibrium.