Overnight Returns: Risk or Conspiracy?
or both?
TL;DR
Virtually all of crypto returns come outside of NYSE trading hours, more so for coins pulled from the top 100, more so than for ETH & BTC
Overnight returns dominate the WallStreetBets meme stock pumps of 2021
This pattern could be a signature of a conspiratorial pump or the nature of risky asset returns
The equity overnight puzzle refers to the fact that, since we had good data on closing stock prices, the overnight returns for most developed countries accounted for all the total equity returns. These are the returns from the NYSE close at 4 pm to the NYSE open at 9:30 am. That diminished substantially around 2010 when 24-hour trading became more common, and it was never large enough to generate a strategy the way, say, weekly mean-reversion in correlated stocks (Coke & Pepsi) worked in the 1990s. Still, it’s interesting to think about what drove that pattern.
Given crypto has no formal trading hours like equities, it was surprising to see the overnight return dominance in the big dogs in crypto—ETH and BTC—given they do not directly have much to do with US equities.
I am bullish on crypto and think it will prove quite valuable in the future when the developed countries all join in unison to deflate their way out of their accumulated debts. The status quo of 5% GDP deficits and expanding off-balance sheet liabilities (eg, social security) is unsustainable, and printing your way out of default politically dominates explicit default. More importantly, the Deep State everywhere would love to implement a social credit score system as in China, freeze fiat assets as Canada did when people protested against their Covid lockdown policy, and demand you justify why you need to withdraw more than $1000 cash from your bank. Crypto will be essential in securing liberty in the future.
That said, anyone who has taken a deep dive into various protocols should know that most crypto coins out there are worthless. Given the cost of spinning up a coin is near zero, and most crypto users are focused on getting rich rather than decentralization, we should expect many worthless imitators. Memecoins and NFTs are transparently worthless, but one could add at least 90% of the rest, as most dapps have no realistic road to decentralization, profitable liquidity providers, or a mechanism for generating real dividends (printing tokens as rewards for staking is what my mentor Hyman Minsky would describe as Ponzi finance).
How do crypto promoters turn worthless coins into billions of dollars? First, it helps if one is stupid enough to think that an NFT with zero IP will have value in the long run; it’s not a lie if you believe it. Another big group consists of amoral marketing mavens. Mike Benz has documented how US intelligence agencies promote their agendas covertly through strategic influence operations, and the same tactics that work for the Deep State work for old-fashioned pump-and-dump scammers. These include paying influencers, creating troll farms to amplify buzz, or funding academia, consulting, and the media to provide favorable ‘independent’ support.
Then there is the direct price manipulation, and perhaps a key tactic is taking advantage of the relative illiquidity when US markets are closed. I pulled data on 55 different coins (non-stablecoins) and looked at their overnight returns.1 Day returns are for NYSE open at 9:30 am to close at 4:00 pm; night returns are from NYSE close to the next NYSE open (they thus include weekend returns). As these coins had various historical lengths (data are pulled from Coinbase), I subtracted the contemporaneous BTC returns to generate normalized coin returns (otherwise, the data picks up whether they missed the recent bear market, etc.). Bitcoin dominates crypto, and it represents a decent crypto market proxy, like the SPY in equities. The relative return disparity was almost exactly like that for BTC, just lower, befitting the fact that BTC has outperformed in this period. Thus, while BTC on average generates zero returns intraday, the average alt-coin generates negative returns intraday (note: these were taken from the current top 100 coins by market cap, which should bias their returns upwards).
As there were many booms and busts among these coins, I defined a bull and bear period for each of these coins by isolating their particular bull and bear markets in this sample period, from January 2020 through July 2025. I took the date of the coin’s maximum price as the peak, and defined its bear market as lasting until it hit its subsequent minimum price. I pulled my list from the top 100 current coins, so there is survivorship bias, but every coin on the list had its subsequent minimum price before the end of my sample period. I defined a coin’s bull market by taking the data from the date of its minimum price before the date of its maximum price.
For these coins, the night returns lead the way up and down. As a comparison, I used March 2020 through November 2021 as the bull market, and November 2021 through December 2022 as the bear market, for both ETH and BTC. These non-scam coins also generated more disparity in their night returns in their various bull and bear markets. That is, the overnight returns dominate the bull and bear movements. Here, I did not adjust the ETH return by the BTC return.
If you float a worthless coin with a fully diluted market cap of $5B, the goal is to sell your insider stake at its maximum price. You know it will be worth zero in 10 years, if not 2. As few coins trade more than 0.5% of market cap a day, cashing out without driving your coin price to zero is a non-trivial problem. One tactic would be to use some of your wealth—which you can borrow against on Maker or Aave—to buy the coin at night when there is low liquidity and sell in the day when liquidity is high. Buying $100 at night and selling $120 intraday might be the optimal strategy for maximizing the extraction from the latest Pump coin.
Interestingly, if you regress the coins' returns against the SPY returns, and split your data into groups where the SPY rises and falls, the difference between the ‘SPY up’ and ‘SPY down’ betas is stark. The nighttime ‘spy up’ betas average 3.86, while ‘spy down’ betas average 2.31. highlights a strong negative convexity consistent with the unsustainability of these coins. Equity returns over the risk-free rate since 2017 in the US have generated an average 12.5% annualized return, which is at least 5% above average. With this tailwind, many crypto promoters think their coin has convexity, and like all deceptions, it is built on a half-truth: upside betas are much greater than 1. In the long run, crypto’s negative convexity will push prices down to zero. For most in crypto, the long run doesn't matter, because as one of my least favorite economists noted, by then we are all dead.
The focus on nighttime trading may also underlie the effectiveness of meme stock pumps. In 2021, the WallStreetBets subreddit decided to strike back at The Man by buying several stocks with high levels of institutional short interest. This ultimately transferred billions of dollars from their rabid base to inept CEOs, but, like all radical redistributionists, they meant well. They were pretty effective, pumping GameStop and AMC up 15-fold within six months. Interestingly, virtually all of these price movements occurred overnight, while the cumulative intraday returns were significantly negative.
This was an overt conspiracy to pump a stock, and while we can’t be certain their tactics focused on buying overnight, it seems likely given their demographics. In January 2021, WSBs had around 10 million members, mostly young retail traders. Many used the Robinhood trading platform, which back then allowed after-hours trading for a modest extra fee of $5, and given most WSB traders had regular jobs, it makes sense they would be trading then as opposed to during market hours.
There were stocks other than GME and AMC affected by the WSB meme-stock bubble. All except one generated gains at night, losses intraday, but all of them had drastically higher returns at night.2
To put this into context, we can look at other periods for these meme stocks. It’s helpful to remember that WSBs targeted these stocks because they were popular shorts, and institutions target shorts based on their fundamental weakness. In a market rally when there was no conspiracy, the 2020 COVID rebound from March 24, 2020, through the end of 2020, these stocks also generated significantly higher nighttime returns. Perhaps the explicit pump returns amplify the return pattern in the absence of a conspiracy. The 2021 pump returns look like the 2020 rally returns times 2 or 3.
Another way to provide context is to look at non-meme stocks. I took the top 100 market cap stocks at the end of 2023 (which could bias my pre-2023 returns, but good enough for a blog post). This provides a baseline for the meme stocks. These regular stocks also generated higher returns overnight in bull markets. However, the disparity is not as striking as it is for the meme stocks, and was not consistently negative in the day.
The meme stocks' rise in their 2021 pump amplified the pattern seen in the 2020 COVID market rebound, and regular stocks also generated higher nighttime returns in bull markets. Similarly, the 50 coins that we can stereotype as worthless amplified the BTC and ETH overnight return disparity.
We know that the meme stock frenzy was a (misguided) conspiracy. We know many crypto coins are worthless, taking advantage of eager, ignorant young men. The significant overnight returns suggest that an efficient pump strategy targets overnight buying and intraday (New York) selling.
On the other hand, the nighttime return disparity could be a benign secondary effect, an epiphenomenon. If and when risky assets rise, their returns could just amplify the day/night return disparity in safe assets. Risky assets could consistently generate negative intraday returns because market makers can’t risk shorting them in the illiquid after-market. If true, the night/day return disparity could be a helpful metric for a risky asset (standard risk metrics include beta, total volatility, negative earnings/assets, average of the top five daily returns over the past month, etc.).
To my low-vol friends, I would suggest sorting stocks by ex ante volatility/beta/etc and looking at the future ‘night minus day’ returns for various deciles (normalized by the absolute value of the average daily return). I would also sort by the ‘night minus day’ returns over the past 6 months or year, and look at the future stock returns of those deciles. Data before 2005 or so probably are not relevant due to the SPY overnight anomaly dying soon afterwards. Still, one needs more than 5 years’ worth of month-ahead returns and at least one bear market to detect anything (thus, I would reach back to capture the 2008 data). It probably won’t help, but it’s worth a shot, as I haven’t seen this metric applied in any empirical research.
Risk, or conspiracy? I wouldn’t be surprised either way.
KAVA, COMP, APE, XCN, HNT, DOGE, MANA, FLOW, ZEC, JTO, SAND, GALA, JASMY, LDO, XTZ, ENS, GRT, IMX, OP, TIA, STX, CRV, INJ, QNT, FIL, FLR, SEI, FET, BONK, ATOM, VET, ARB, ALGO, ICP, APT, ONDO, NEAR, ETC, AAVE, CRO, DOT, UNI, SHIB, LTC, AVAX, HBAR, BCH, LINK, SUI, XLM, ADA, SOL, XRP, REP
BlackBerry – BB ,Koss Corporation – KOSS , Express, Inc. – EXPR , Nokia – NOK , Clover Health – CLOV , Wish – WISH , Tupperware – TUP