Tradiens
EssaysSign in

Tradiens Research · Data essay · Equities

The Night ShiftFrom 1993 to 2020, large US stocks earned their gains while the exchange was closed.

It seems obvious that stocks make their money while the market is open, because that is when the exchange is trading and the price can be seen to move. For SPY, a fund that tracks the 500 large US companies of the S&P 500 index, the record from 1993 to 2020 says something else: the gain came between the evening close and the next morning's open, and the trading hours, taken together, took part of it back. If the return arrived at night, why could nobody simply own stocks at night and collect it?

Version 2.0First published Sep 26, 2026Last updated Oct 3, 2026

Scroll to see the four ways of holding the fund

A 2008 study

A working paper dated 2008 found that, over the decade before it, the extra return on US stocks had come entirely while the market was closed.

In a working paper dated 2008, three finance researchers, Michael Cliff, Michael Cooper and Huseyin Gulen, looked at the equity premium, the extra return that stocks pay above a safe investment as the reward for bearing their risk. A working paper is a study shared with other researchers before a journal has reviewed it. Their finding, in their own words, was that "the US equity premium over the last decade is solely due to overnight returns", while returns during the day were "close to zero and sometimes negative", over a decade that the paper's summary does not date.

A return can belong to the night because every trading day can be split into two parts. The overnight part runs from one day's closing price to the next day's opening price, and the daytime part runs from that opening price to the close. Because the daytime part starts exactly where the overnight part ends, an owner who holds the share through both receives both, and the two multiply into the result of the whole day. That means that when two of the three results are known, the third follows from them.

Twenty-seven years of one fund

From 1993 to 2020, holding the S&P 500 fund SPY only overnight would have gained more than holding it all the time.

A decade is a short record, and a 27-year record of the same split exists for SPY. SPY is the SPDR S&P 500 ETF, an exchange-traded fund, which is a fund whose shares are bought and sold on the exchange like a stock, so its price has an open and a close every trading day. In 2020 Matthew Bartolini, head of SPDR Americas Research at State Street Global Advisors, the firm that issues SPY, split the fund's record from January 1993 to January 2020 into its nights and its days, in a guest post on the blog of the investment firm Alpha Architect.

SPDR S&P 500 ETF (SPY), January 1993 – January 2020 · price return, dividends not included · before trading costs · SPDR Americas Research, via Alpha Architect (2020)

Holding SPY only overnight would have beaten holding it all the time.

Each cell starts from $1. "Overnight" means buying at the close and selling at the next morning's open.

Nights only × days only = all the time: 8.17 × 0.89 = 7.27. Each tick doubles the money.

Every bar starts at $1. A solid bar is the source's figure; an outlined bar is our calculation from the source's two figures (7.27 ÷ 8.17 − 1, price return, before costs).

The two ways of owning the fund came out in the opposite order from the one most owners would expect. Every bar in the table starts from $1; compare the "Nights only" cell with "All the time". Someone who had bought at every close and sold at every next morning's open would have gained 717%, so that each dollar became $8.17, while someone who had bought once and simply held would have gained 627%, each dollar becoming $7.27. Both are counted on the price alone and before trading costs.

The days

Now look at "Days only", whose bar is outlined because it is our calculation, not the source's figure: it is worked out from the other two, as the line under the table shows (8.17 × 0.89 = 7.27). Because the nights alone came out ahead of the whole, the days, taken together, can only have subtracted. Dividing what holding produced by what the nights produced leaves the days with a loss of about 11% (our calculation from the source's two figures; price return, before costs). The source also prints a daytime figure of its own, but its three figures do not multiply together, so we leave that one out.

Researchers have proposed two partial explanations for why the trading hours did worse than the nights, an opening price pushed too high in much-watched single stocks and a discount taken at the close in the market as a whole, and neither of them puts a number on SPY's daytime loss.

It would be easy to conclude from this that the night won almost every day, and the day-by-day record shows how far that is from what happened.

Day by day

The night beat the day on only 53 days in 100, so the large total came from a small edge repeated thousands of times.

Bartolini also compared each night with the trading day that followed it, so that the move from Monday's close to Tuesday's open was set against the move from Tuesday's open to Tuesday's close. On that comparison the night came out ahead on 53% of the roughly 6,800 trading days in the record, which is only a little more often than it came out behind. The night was therefore not a reliable winner from one day to the next.

The difference on a typical day was also tiny: the median night gained 0.05%, and the median day lost 0.04%. A difference that small still grows large over time, because each day's result is applied to everything that the days before have left, so the small edge is multiplied day after day. Repeated over 27 years of trading days, it became the gap between the two totals. The surprise lies in that compounding, and not in how often the night won. Compounding left the total, however, no more certain than the daily differences it was built from, and the source's own statistical test could not quite rule out chance as the explanation for the average daily difference between night and day.

The edge also weakened within the window. Up to the end of 2002, which the source takes as the end of the dot-com bubble, the night-only strategy beat simply holding in nearly two-thirds of all one-year stretches, whatever day they began, and afterwards in only about a third, so the night's advantage was already fading in the 2000s. In the index the advantage was small and uneven, while a single stock shows how far the same split could go.

One stock

On AMC's shares, buying at every open and selling at every close would have lost 99.6%, while holding only overnight would have turned each dollar into about $300.

In June 2022 Victor Haghani, the founder of the investment firm Elm Wealth, published the same split for single stocks with Vladimir Ragulin and Richard Dewey, and one of their cases was the cinema chain AMC. Their AMC window runs from the start of 2019 to late May 2022.

Over those three and a half years, someone who had bought AMC at every open and sold at every close would have lost 99.6% of the money. Someone who had held it only overnight would have gained 30,000%, which turns each dollar into $301.

AMC Entertainment, start of 2019 – late May 2022 · before trading costs · Elm Wealth, "Night Moves" (2022)

On AMC, day-only holding would have lost 99.6%; night-only would have turned each dollar into about $300.

Each tick is ten times the one to its left. A gain of 30,000% turns $1 into $301 (the dollar plus 300 more). The bars show where $1 ended, not a path over time. The source's text gives no figure for holding AMC all day and night in this window.

A study published ten years earlier helps explain how a much-watched stock can lose by day what it gains by night. Berkman, Koch, Tuttle and Zhang examined the largest US stocks from 1996 to 2008 and found that buying by individual investors in stocks that had caught their attention was concentrated near the open, which pushed the opening price high compared with the rest of the day. For the most-watched stocks, the average night gained 13 basis points, or 0.13%, and the average day lost the same amount.

The Elm authors take this as the most plausible explanation for stocks like AMC. In their view, small investors decide what to trade while the market is closed and place their orders at the open, which pushes the opening price up. If that is right, an opening price that is too high makes the night look better and the following day look worse by the same amount, without making anyone richer, which would push the two AMC figures far apart.

The Elm authors argue that this effect could reach the whole index too, through the weight of the most-watched large companies in it. For the index, however, a study published by the Federal Reserve Bank of New York proposes a different explanation, built around a single hour.

One hour

In S&P 500 futures, almost all of the night's return came in a single hour, between 2 and 3 a.m. New York time.

To see where in the night the return arrived, Boyarchenko, Larsen and Whelan, in a Staff Report of the Federal Reserve Bank of New York, turned to S&P 500 futures, the contracts known as E-minis. A futures contract of this kind is an agreement on the value of the index at a set future date, settled in cash, and these contracts have traded close to 24 hours a day since the late 1990s, so their price can be followed hour by hour. Their record runs from 1998 to 2020.

The hour from 2 to 3 a.m. New York time earned an average of 3.7% a year, and the whole night earned 3.6%. One hour can match the whole night because the other hours of the night, some up and some down, came to about zero together. These are yearly averages measured on the futures, and in a different way from the SPY totals.

Measured against the whole day, from one close to the next, which earned 5.9% a year, that hour supplied more than 60% of the total. It was no accident of a few good years either, because the hour was positive in 20 of the 23 years of the record.

S&P 500 E-mini futures, 1998–2020, 5,691 trading days · average yearly return of the futures, before costs · New York Fed Staff Report 917

One hour of the twenty-four earned as much as the whole night together.

All three bars share one scale. The other night hours together came to about zero, which is why one hour can match the whole night. The study also lists the average for every other hour; most are small, and some are negative.

The authors' explanation, which they put forward as an argument, starts at the close. When sellers outnumber buyers late in the day, dealers, the trading firms that stand ready to buy and sell, take the unwanted contracts off their hands, but only at a price lowered enough to pay them for holding through the night; that lowered price becomes the day's close, so the day gives up a little and the night gets it back when the price recovers. That fits what they found: after a day of heavy selling the night's rebound was strong, and after a rise it was much smaller. The mechanism moves part of the return from the day into the night, but it does not make the day a loss, since on these futures the trading hours still gained about 2.3% a year, the whole day's 5.9% less the night's 3.6%.

For the timing after a sell-off, the authors offer a clue rather than a proof. The hour falls around the time European trading begins, they note, when volume jumps, and they suggest that this is when the dealers who took on the evening's unwanted positions can pass them on to new buyers, so that is when the price recovers.

The cost of trading twice a day

Counted with the trading costs of the time, the same overnight strategy ended at −32% instead of +717%.

Knowing when the return arrived suggests a plan, which is to own the market only at night, and Bartolini tested it on SPY with the rule of buying at every close and selling at every open. That meant two orders every trading day, or about 500 a year. Each order paid twice over: a commission to the broker, and the cost of crossing the bid-ask spread, which is the gap between the price at which you can buy at a given moment and the slightly lower price at which you can sell.

The test charged the bid-ask spread and the commission in full. The purchase price was raised, and the sale price lowered, by the whole spread of that day and by an estimated commission of one cent per share. With the trading costs of the time, the overnight strategy that had gained 717% ended 32% below its starting point. This is a backtest, a calculation run on past prices, and not a loss that anyone booked.

SPDR S&P 500 ETF (SPY), January 1993 – January 2020 · price return, dividends not included · before trading costs unless marked · SPDR Americas Research, via Alpha Architect (2020)

Counted with the trading costs of the time, nights only would have ended at $0.68.

Each cell starts from $1. "Overnight" means buying at the close and selling at the next morning's open.

Nights only × days only = all the time: 8.17 × 0.89 = 7.27. Each tick doubles the money.

Every bar starts at $1. A solid bar is the source's figure; an outlined bar is our calculation from the source's two figures; the second bar under "Nights only" is the source's backtest with the trading costs of the time, not a loss anyone booked.

The cost won because it was about the size of the prize and was paid every day. Working back from the source's two end results, our calculation puts the average cost of a night's two trades slightly above what an average night earned. A cost slightly larger than the average gain, paid every trading day, compounds into a loss in the same way that the small edge compounded into a large one.

Because the source used the spreads of its own window, with cost rules that could err in either direction, its result cannot be carried over to the cost of trading today.

A test with different assumptions, on a different instrument and window, points the same way. For 2004 to 2020, the New York Fed study's authors calculated a strategy that bought the E-mini at 2 a.m. and sold it at 3 a.m. every trading day. On paper, it was profitable on average before costs and loss-making on average once the bid-ask spread was paid. Their table mentions no commission, which a real trader would also have paid.

That was a calculation. In 2022 someone tried the idea with real money.

A fund that tried

One fund held the S&P 500 only at night with real money, and it closed after thirteen months.

The NightShares 500 ETF, with the ticker NSPY, began trading in June 2022, and its investment adviser was AlphaTrAI Funds. It offered the overnight idea to the public as a fund. During the day it kept its money in cash and short-term US government debt, and in the evening it bought S&P 500 futures, so that at night it moved almost one-for-one with the index and by day stood outside the stock market.

Its annual report to the Securities and Exchange Commission, the US market regulator, covers only its first eleven months, to the end of May 2023. In that time the fund lost 7.20% after its fees while the S&P 500 index gained 11.15%, so it trailed the market it followed by more than 18 percentage points.

The managers' own account in the same report explains where the gap came from. They had launched the fund to capture the night effect, which they called "a well-researched source of outperformance", and in their reading the year went the other way: the daytime markets were strong while the overnight sessions were negative overall, and their fund, by design, missed the daytime gains.

In July 2023 the fund's board decided to close it on AlphaTrAI's recommendation, and the filings give no other reason. Its last trading day was July 31, thirteen months after launch, after which the fund was to be wound up and its remaining money paid out to its shareholders. A second NightShares fund, built to hold more of the index by night than by day, closed a month later.

NightShares 500 ETF (NSPY) · overnight fund · June 28, 2022 – May 31, 2023 · real money, after fees, dividends reinvested · SEC Form N-CSR

The real overnight fund lost 7.20% in eleven months while the S&P 500 gained 11.15%.

  1. 1June 28, 2022: the fund launched.
  2. 2May 31, 2023: end of the eleven months the −7.20% covers. June 1 – August 10, 2023: no return for this stretch is in the filings we opened.
  3. 3July 14, 2023: the board decided to close the fund, on the adviser's recommendation.
  4. 4July 31, 2023: last trading day.
  5. 5On or about August 10, 2023: due to be wound up and the money paid out (open ring: a planned date).

Both bars are from the fund's annual report, for the same eleven months.

How much of that shortfall came from costs, and how much from the pattern turning, is not known. The result includes the interest the fund earned on that government debt as well as its fees and trading costs, and the filings separate neither those items nor the two causes. Nor is the fund a test of the backtest before it, because it traded a different instrument, in different years, at different costs.

The authors of the New York Fed study later measured what happened to their hour across the years around the fund, from 2021 to 2025.

After 2020

The authors of the New York Fed study report that the 2-to-3 a.m. hour returned close to zero from 2021 to 2025.

In July 2026 the same authors returned to the hour in a post on the New York Fed's blog, Liberty Street Economics, covering January 2021 to December 2025. They found that the return between 2 and 3 a.m. had been "close to zero" over that period, and that the same was true of the futures on the Nasdaq-100 and the Dow. Looking back, they wrote that the two NightShares funds closed in a way consistent with the weaker pattern they measured after 2020, though costs specific to the funds likely also played a part.

Their explanation follows from the first one. In the last hour before the close, the imbalance between sellers and buyers varied less than half as much from day to day as before. As a result, dealers had less to take on at the close and earned less for taking it. The two other things their explanation could point to, the price swings the market expected and the share of futures trading done at night, had stayed about the same, so of the inputs they could measure only the imbalance at the close had changed much, an explanation they still call work in progress.

Their update covers only that hour and gives no figure for the whole night, so it does not show that the overnight return disappeared, and for SPY a different source suggests that it did not. In the five years to March 2025, by the Elm authors' count, a dollar held only overnight grew to $1.66 and a dollar held only by day to $1.25. The two findings can both be true, because one hour of futures trading going flat leaves the other hours of the night free to pay, and the update gives no figure for them. That leaves the question of who, in the end, kept the night's gain.

Who kept the gain

Whoever simply held the fund collected the night's gains along with the days, and paid for no extra trades.

Owning only the night did not pay in this record, because at the costs of the time, as the source counted them, trading cost on average slightly more than the night paid. The fund that tried it with real money on the S&P 500 also met a year in which, by its managers' account, the days were strong and the nights weak. The gain went instead to the investor who never tried to separate the night from the day. Holding SPY for the whole twenty-seven years earned less than the nights alone did on paper, because the holder also carried every trading day, but the holder paid for no extra trades and did not have to guess how long the pattern would last.

The comparison is fair only within each window. Over the same years, the overnight strategy ended below its starting point once its trading was paid for, while the holder ended with more than seven times the money. Over the fund's eleven months, the fund lost money while the index rose. Neither comparison is a forecast of what the night will pay from now on.

The gain arrived at night, but in this record the cheapest way to collect it was to sit through the day as well.

Sources & method

This essay brings together six groups of figures, each with its own window: SPY's price return from January 1993 to January 2020, without dividends, from a 2020 analysis by SPDR Americas Research; the realized, after-fee return of the NightShares 500 ETF from its annual report to the SEC, June 28, 2022 to May 31, 2023; hourly returns of the S&P 500 E-mini futures from 1998 to 2020, from a New York Fed Staff Report, and its authors' update for 2021 to 2025; the overnight and daytime returns of single stocks and of SPY from two essays by Elm Wealth; and the arguments of five studies about possible causes. We did not compute any return series ourselves, because we found no daily open-and-close series that we were permitted to republish, so every figure shows named endpoints and no lines between them.

The two halves of a day.
Each trading day splits into an overnight return (close to next open) and a daytime return (open to close). When both use the same opening price, (1 + overnight) × (1 + daytime) = (1 + close-to-close), so the two multiply into the return of holding.
The daytime figure.
The source reports +717% for nights only, +627% for holding, and +12% for days only. The three cannot all be right: 8.17 × 1.12 = 9.15, not 7.27. The daytime multiplier implied by the other two is 7.27 ÷ 8.17 = 0.890, a loss of about 11%; that is the figure in the essay, labelled as our calculation. It rests on the source's two figures being correct and computed from the same opening prices. It does not depend on dividends, because a dividend is taken off the price at the open, which falls in the overnight part. We do not say why the source's daytime figure differs; we only do not use it.
The overnight strategy with costs (−32%).
A backtest by the source: the purchase price raised and the sale price lowered by the full bid-ask spread of that day and by an estimated commission of $0.01 per share; trades at the closing and opening levels; no slippage. 6,800 trading days over 27.0 years is about 252 round trips, or 504 orders, a year. Our calculation, as geometric averages per round trip: average overnight gain ln(8.17) / 6,800 = 3.09 basis points; average cost needed to reach −32%, ln(8.17 / 0.68) / 6,800 = 3.66 basis points (3.07–3.11 and 3.63–3.68 if the day count is anywhere from 6,750 to 6,849). They restate the source's two figures and are not independent evidence. The source does not say how it measured the day's spread. Its cost rules can err in both directions: Elm Wealth notes that trades in the opening and closing auctions, when the exchange matches all waiting orders at a single price, do not cross a bid-ask spread, which would make the source's cost too high; charging the whole spread on each side, rather than the usual half spread measured from the midpoint, also leans toward too high a cost; leaving out slippage would make it too low; which is larger cannot be told. The result uses the spreads of its window and does not describe the cost of trading today. The author was head of SPDR Americas Research at State Street Global Advisors, the issuer of SPY.
Day by day.
The 53% compares each overnight return with the daytime return of the day that follows it. The paired t-test on the two daily return series gave t = 1.90 and p = 0.06, against the source's 5% level; it tests the mean daily difference, not the median (+0.05% overnight, −0.04% daytime) and not the 53%. Up to December 31, 2002, which the source takes as the end of the dot-com bubble, buying at the close and selling at the open beat holding in 64% of rolling one-year and 61% of rolling three-year periods; after it, in 34% and 32%.
The single hour.
New York Fed Staff Report 917 (at the time of the Staff Report, 2022: Boyarchenko of the Federal Reserve Bank of New York; Larsen and Whelan of Copenhagen Business School): the most liquid E-mini S&P 500 contract, almost always the front month; returns from the midpoint of the best bid and offer; hour-by-hour log returns, annualized (1.48 basis points a day for the 2–3 a.m. hour, about 3.7% a year); January 5, 1998 to December 31, 2020, 5,691 trading days; the night is 4:15 p.m. to 9:30 a.m. Eastern time. The authors compare their 5.9% for the whole day with the S&P 500 cash index's log return excluding dividends. "More than 60%" is their wording; 3.7 / 5.9 = 63% is our arithmetic. The returns are log returns, which add across hours of the same days; the other night hours together came to about zero (3.6 − 3.7 lies within rounding of zero). The study's Table I lists the average of every one of the 24 clock hours (basis points a day): most are small and some negative; the 24 hours sum to 2.34 basis points a day, which × 252 = 5.90% a year. Those clock-hour bands cannot be reconciled with the study's 4:15 p.m.–9:30 a.m. night, so the 3.6% night is not rebuilt from them and the hourly profile is not drawn. On the authors' inventory argument the discount at the close moves part of the return from the day into the night; it does not make the day a loss, since on these futures the trading hours gained about 2.3% a year (our subtraction of two rounded, annualised log returns on the futures, 5.9 − 3.6; it describes the futures, not SPY, and is uncertain by about ±0.1), and it is not offered as the cause of SPY's negative days in a different window and measure. The timing argument: at 2:00 Eastern time regular trading opens on Euronext, Eurex and the Frankfurt exchange, with pre-market trading in London, and volume jumps; the authors call this "a first clue towards a potential explanation". The cost test (Table IX) runs from January 2004 to December 2020, because the overnight spread reached its minimum of one tick in 2004: the 2–3 a.m. strategy averaged 3.75% a year before costs and −0.59% after paying the bid-ask spread (Sharpe ratio 1.10 and −0.54, with the 4-week Treasury bill as the risk-free rate); the table mentions no commission.
After 2020.
Liberty Street Economics, July 1, 2026: January 2021 to December 2025, 1,245 trading days. The 2–3 a.m. return is described as "close to zero" and "flat" with no figure. The standard deviation of the last-hour signed volume measure fell from 6.5 to 2.9; the VIX, an index of the price swings the market expects, averaged 20.4 and 19.4 (medians 18.6 and 18.2); night trading was 15% and 16% of E-mini volume. The update covers only the 2–3 a.m. hour. The authors call the explanation preliminary. They report the VIX and the night share of volume as the measures of the two other channels their model names, both "roughly unchanged"; of the measures available, only the closing imbalance "shifted substantially". They write that both NightShares funds closed "consistent with the weakened pattern over this sample (though ETF-specific costs likely also contributed)"; their "fourteen months" is not used, because the SEC dates give thirteen.
NightShares.
Investment adviser: AlphaTrAI Funds, Inc. NSPY: launched June 28, 2022; held cash and/or US Treasuries during the day (on May 31, 2023, a six-month target-duration Treasury ETF made up 67.55% of net assets) and 17 long E-mini S&P 500 contracts with a notional value of $3,561,925 against net assets of $3,627,050; NAV return −7.20% and market-price return −7.35% to May 31, 2023, after fees, dividends reinvested; S&P 500 +11.15% (total return); net expense ratio 0.55% a year, excluding trading and futures commissions; the board determined on July 14, 2023 that closing the fund was in the shareholders' best interests "following a recommendation by the Funds' investment adviser", and the filings give no other reason; last trading day July 31, 2023 (398 days after launch); due to be liquidated on or about August 10, 2023. NSPL was designed to hold the S&P 500 at 1× by day and 1.5× at night; launched October 4, 2022; +10.20% against +11.54% to May 31, 2023; expense ratio per the prospectus 0.67% after waivers (0.77% before); decision August 14, 2023, last trading day August 31, 2023, due to be liquidated on or about September 8, 2023. The managers' discussion in the N-CSR is qualitative and quarter by quarter (no night and day figures): they launched the funds to capitalize on "the strength of the night effect, a well-researched source of outperformance"; a "bear market rally" in the daytime session in July–August 2022; in the first quarter of 2023 a market "again led by the daytime session" with the night negative; the shortfall "a function of strong day time markets alongside negative returns overall in the overnight session"; and "20 years of history (and numerous academic studies) have shown that the Night sessions can provide strong returns". All of it is the managers' reading. The fund's final return is not in the filings we opened. The filings do not separate the effect of costs from the effect of the pattern's reversal.
Single stocks.
Elm Wealth (Victor Haghani, founder; with Vladimir Ragulin, a consultant to Elm, and Richard Dewey), "Night Moves" (June 20, 2022): AMC from the start of 2019 to late May 2022, before costs; exact days not given. AMC describes itself in its annual report for 2021 as "the world's largest theatrical exhibition company". "Still Working the Night Shift" (March 3, 2025; Haghani, White, Dewey and Ragulin): SPY over the five years to publication, $1 → $2.07 held throughout, $1.25 day only, $1.66 night only (1.25 × 1.66 = 2.075); Yahoo Finance data; dividend treatment not stated.
Causes.
Every explanation in the essay is its authors' argument and none is settled: the inventory-risk and European-open argument of Boyarchenko, Larsen and Whelan ("we argue", "a first clue"); the attention finding of Berkman, Koch, Tuttle and Zhang (the 3,000 largest US stocks, 1996–2008: for high-attention stocks +13 basis points a night and −13 a day, for low-attention stocks +3 and −3; the hidden cost to retail traders who buy high-attention stocks near the open "frequently exceed[s] the effective half spread"); Elm Wealth's view that retail buying at the open "seems the most plausible" explanation and could reach the index through large high-attention stocks; Bogousslavsky, who argues that higher overnight margin requirements and lending fees typically charged only on positions held overnight give arbitrageurs a reason to reduce their positions before the close, and finds that a mispricing factor does poorly at the end of the day (this concerns end-of-day distortion, not the index's overnight return); Qiao and Dam on China's rule that shares bought on one day cannot be sold until the next (often labelled "T+1", which in the US usually means the settlement cycle instead).
Color.
Every figure shows percentages or growth of $1, which are ratios, so none uses the money color.

What we left out, and why

How much of the NightShares fund's shortfall came from costs and how much from the reversal: no source separates them. The fund's return over its whole life. The whole-night return after 2020: the update covers only the single hour. The years covered by the 2008 working paper: its summary gives none, and the full text could not be opened. Crypto markets, which never close: we found no primary source for how the split between day and night behaves there, and our own intraday data have not been verified for this use. Daily lines for SPY: we found no series we may republish. The size of China's overnight discount: the one estimate we found has no stated window. The sister fund NSPL in detail, a second AMC window and GameStop (Elm, 2025), a 30-year SPY figure from a different source and return type, and a study of trading-strategy returns (Lou, Polk and Skouras, 2019): they are in this note, repeat the story, or answer a different question.

References

  1. Cliff, M. T., Cooper, M. J., & Gulen, H. (2008). Return differences between trading and non-trading hours: Like night and day (SSRN Working Paper No. 1004081). archived abstract page
  2. Bartolini, M. (2020, June 16). Trading costs wipe out the overnight return anomaly [Guest post]. Alpha Architect. Data: Bloomberg Finance L.P., calculations by SPDR Americas Research. archived copy
  3. Boyarchenko, N., Larsen, L. C., & Whelan, P. (2020, rev. 2022). The overnight drift (Federal Reserve Bank of New York Staff Report No. 917); published in Review of Financial Studies, 36(9), 3502–3547 (2023). newyorkfed.org
  4. Boyarchenko, N., Larsen, L. C., & Whelan, P. (2026, July 1). The disappearing overnight drift. Liberty Street Economics, Federal Reserve Bank of New York. libertystreeteconomics.newyorkfed.org
  5. Unified Series Trust. (2023). Form N-CSR, annual report for the period ended May 31, 2023 (NightShares 500 ETF, NightShares 500 1x/1.5x ETF), accession 0001580642-23-004064; Form 497 supplements of July 14, 2023 (0001580642-23-003676) and August 14, 2023 (0001580642-23-004244). sec.gov (N-CSR) · sec.gov (497, July 14)
  6. Haghani, V., Ragulin, V., & Dewey, R. (2022, June 20). Night moves: Is the overnight drift the grandmother of all market anomalies? Elm Wealth. elmwealth.com
  7. Haghani, V., White, J., Dewey, R., & Ragulin, V. (2025, March 3). Still working the night shift. Elm Wealth. elmwealth.com
  8. Berkman, H., Koch, P. D., Tuttle, L., & Zhang, Y. J. (2012). Paying attention: Overnight returns and the hidden cost of buying at the open. Journal of Financial and Quantitative Analysis, 47(4), 715–741.
  9. Bogousslavsky, V. (2021). The cross-section of intraday and overnight returns. Journal of Financial Economics, 141(1), 172–194. ideas.repec.org
  10. Qiao, K., & Dam, L. (2020). The overnight return puzzle and the "T+1" trading rule in Chinese stock markets. Journal of Financial Markets, 50. ideas.repec.org
  11. AMC Entertainment Holdings, Inc. (2022). Form 10-K for the year ended December 31, 2021, accession 0001558370-22-002577.

Limits

A record of past windows, each stated, and not a forecast, a strategy or advice. The essay carries no live data.

How to cite this essay

Tradiens Research. (2026). The night shift (Version 2.0). Tradiens Data Essays. https://tradiens.com/essays/night-shift

@misc{tradiens2026nightshift,
  author       = {{Tradiens Research}},
  title        = {The Night Shift},
  year         = {2026},
  version      = {2.0},
  howpublished = {Tradiens Data Essays},
  url          = {https://tradiens.com/essays/night-shift}
}
Tradiens Research · Data essay · v2.0 Trade like you know.