KOSPI 200 Night Futures

A real-time chart of KOSPI 200 futures trades in the night session, from 18:00 to 06:00 KST the next morning, stacked into 1-minute bars. The reference line is the previous regular-session close, and the figures are futures prices, not the index.

Night futures are index futures traded in the extended session that runs from 18:00 to 06:00 KST once the regular session has closed, and their underlyings are KOSPI 200, built from large caps, and KOSDAQ 150, weighted toward growth technology names. Both contracts are listed on a quarterly cycle, trade for the last time on the second Thursday of the expiry month and trading migrates to the following quarter as expiry approaches, so what the screen draws is always the nearest contract. The regular session opens at 09:00 and closes at 15:45, leaving a little over two hours of dead time between the two sessions. When the Korea Exchange brought the night derivatives market onto its own platform the range of eligible products widened and night trading in KOSDAQ 150 began at that point, so the samples the two contracts have accumulated differ greatly in length.

The case for an extended session rests on the fact that price-relevant information does not arrive only during domestic trading hours. Eun and Shim (1989) found the flow of innovations from the United States market to others dominant over the reverse direction. That volatility as well as returns crosses borders was established for New York and Tokyo by Hamao, Masulis, and Ng (1990). Lin, Engle, and Ito (1994) set out the structure in which one market's day meshes with another's night, and Mun and Hong (2003), on spillover from NASDAQ 100 into KOSDAQ 50 index futures, is an early case confirming that the same route runs into Korea. Where trading hours are confined to the day this information waits for the open, and the extended session is the device that shortens the wait. That frictions in the regular session changed once night trading began was measured on KOSPI 200 futures by Seon and Joo (2019).

A futures contract's fair value follows from the cost-of-carry relation (Cornell and French 1983).

Here $S_{t}$ is the level of the underlying index at the previous regular close, $r_{t}$ the risk-free rate, $d_{t}$ the expected dividend yield over the remaining life and $\tau_{t}$ the time to expiry measured in years. How far an observed price stands from that value is the fair-value deviation.

Here $F_{t}$ is the futures price observed in the night session, and the gap $b_{t}$ carries transaction costs, short-sale constraints, uncertainty about dividends and the net demand of hedgers and speculators all at once. At night the cash index is frozen at the regular close and $S_{t}$ does not update, so price movement in these hours is for the most part a revision of expectations about the next open. The finding of Nam et al. (2006) that futures impound information ahead of the cash market was obtained in the regular session, but at night, when the cash market does not trade, the futures price is the only market price attached to the value of the underlying.

The night book is thinner than the regular session's. The combination Barclay and Hendershott (2003) report for after-hours trading, low volume and wide spreads alongside a larger share of the price move carried by each trade, was reported for the United States after-hours market, and in such a market it separates the levels at which a bar can be read. At tick level it is easy to misread a wick produced by a handful of trades as information, whereas the move across a whole session is closer to a summary of the news that arrived that night. That the price impact of a trade varies with the hour of the day is something Webb et al. (2016) had already shown for the regular session, and as the hour changes a trade of the same size leaves a different mark. Lee and Woo (2013) addressed price-discovery efficiency in the night session head on. There is no reason to assume that the night's mix of participants matches the day's, but none of the studies cited here measures that mix contract by contract. Kang (2021) traced how cross-product trading settled in as the KOSDAQ 150 futures market grew.

The night session is taken as the twelve-hour window that opens at 18:00 and closes at 06:00 KST the next day. Each session is identified by its starting date, that is, by the date of the evening, so a session beginning on a Wednesday evening and ending on a Thursday morning carries the Wednesday date. Sessions run on evenings from Monday through Friday other than market holidays, and one opened on a Friday evening closes on the Saturday morning. Anchoring the label to the starting date keeps a single night from being cut at midnight, and it divides time at the same place as the partition of Lin, Engle, and Ito (1994), which pairs one market's day with another's night.

Trades are assigned to buckets obtained by flooring their second-level timestamps to the minute, and every bucket yields four prices and a volume.

Here $p_{i}$ and $v_{i}$ are the price and quantity of an individual trade in that bucket and $k$ is the bucket's index. The open and the close are the prices of the earliest and the latest trade inside the bucket, and the high and the low are the largest and the smallest within it. A minute holding no trade at all produces no bar. Filling an empty minute with the previous value erases the fact that nothing traded, and the observation of Barclay and Hendershott (2003) that volume and per-trade price contribution move in opposite directions in the after-hours market makes that fact data one cannot discard. The stealth-trading argument of Barclay and Warner (1993), in which the share of cumulative price change depends on trade size, adds a further reason not to synthesize trades, since an interpolated bar is not a bar of zero volume but a trade that never happened.

Higher timeframes are not rebuilt on the server but computed on the screen by grouping the 1-minute bars. Group boundaries are cut without overlap from the session's opening time, and consecutive 1-minute bars are combined as follows.

Here $n$ is the number of minutes grouped, and the chart offers 5-minute, 10-minute, 15-minute and 30-minute groupings, while $B$ is the set of 1-minute bars in the group, $k$ the earliest bar in it and $k'$ the latest. Empty minutes inside a group are still not filled, so the volume of a coarse bar is the sum of quantities actually traded, and the coarser the grouping the more the small movements disappear and only the larger sweep remains. Given the result of Webb et al. (2016) that the price impact of a trade is not constant through the day, a comparison across bar sizes holds only when bars from the same hours are set against each other.

Movement inside a session is measured against the close of the previous regular session.

Here $C_{t}$ is the most recent traded price and $P_{\mathrm{prev}}$ the close of the previous regular session, and the value is shown in percent. The reference sits at that close because the cost-of-carry relation makes the cash level at that same moment the starting point of fair value (Cornell and French 1983), and measured that way the night's movement can be set on the same axis as the daily change quoted during the day.

The quantities built this way lead straight into the measures the literature uses. Treating the stretch from the close to the next open as a single return is the partition that separates overnight from intraday returns, and both the result of Bondarenko and Muravyev (2023) that most of the average daily return in S&P 500 futures falls inside a narrow band outside regular trading hours and the drift that Boyarchenko, Larsen, and Whelan (2020) place around the European open are defined on it. Measuring how much of the price a given stretch of a session moved requires accumulating the contribution of each stretch, and the study that treats price-discovery efficiency in the night session that way is Lee and Woo (2013). The questions asked by Ahn et al. (2024), who set retail flows against the gap between overnight and intraday returns, and by Kwak (2024), who links retail trading intensity to returns, presuppose the same partition. What this screen publishes, however, stops at the bars and the change against the reference price, and price contribution, price impact and estimators of the kind Nam et al. (2006) used to measure how far futures lead the cash market are not computed here.

The night session is the channel through which shocks originating outside regular trading hours reach Korean equity prices. When United States employment figures land around 22:30 KST and a rate decision comes through near 04:00, this session is the only domestic market in which the risk carried by KOSPI 200 can be repriced on the spot. The index is heavy in exporting large caps with substantial foreign ownership, so a shift in global risk appetite reprices its risk even when nothing domestic has changed. The frame for cross-border transmission was laid down long ago. That a country's share prices cannot be explained by its own circumstances alone was already visible in Eun and Shim (1989), and Hamao, Masulis, and Ng (1990) added that the spillover reaches volatility as well as returns. The partition of Lin, Engle, and Ito (1994), which cuts night and day into pairs, gave the channel a frame in which it could be measured.

The evidence built on Korean data is concentrated on the KOSPI 200 contract. Lee and Woo (2013) measured the efficiency of price discovery in the night session, and Yeom, Baek, and Ryu (2013) examined the performance of investment strategies that use this contract. Seon and Joo (2019) measured how frictions in the regular session changed once night trading began, a question motivated by the possibility that informed trading which had concentrated in the opening call spreads into the night as the non-trading interval shortens. Information arriving earlier and information coming into existence are different events, however. Set beside the relation of Nam et al. (2006), in which futures impound information ahead of the cash market, the night session is a device that shortens the time information spends waiting for the open.

The observation that returns concentrate outside regular trading hours is not confined to Korea. Bondarenko and Muravyev (2023) showed that most of the average daily return in S&P 500 futures falls inside a narrow band, and Boyarchenko, Larsen, and Whelan (2020) recorded that the concentration sits around the European open. Translated into Korean time, that moment falls close to the dead stretch just after the regular market closes and before the night session opens, and what the night session covers is instead the hours when the United States regular market is trading. Both results were obtained on United States index futures, so where information concentrates within the KOSPI 200 night session does not follow from them directly.

The mix of participants also changes the character of the channel. Kwak (2024), linking Korean equity returns to the trading intensity of individual investors, and Ahn et al. (2024), setting retail flows against the gap between overnight and intraday returns, point the same way, namely that night and day returns are formed under different participant mixes. That is also why a relation observed during the day does not transfer to the night unchanged. The observation of Barclay and Hendershott (2003) that after-hours volume is low while the price contribution of each trade is large says that fewer participants is not the same as less information. The comparison of Yoon (2021), in which the informational efficiency of KOSPI and KOSDAQ is not the same, shows that the same caution is needed when the underlying index is swapped.

The first use of night futures is to take the edge off the gap that opens between the close and the next open. A fund tracking KOSPI 200, or an investor holding names with a large index weight, carries the risk of the stretch from the 15:30 stock-market close to the 09:00 open the following day, and the only liquid domestic market open in between is night futures. A hedge ratio is obtained by regressing the next morning's opening gap on the night return of the same night.

Here $\Delta S^{\mathrm{open}}_{t}$ is the underlying index return from close to next open, $\Delta F^{\mathrm{night}}_{t}$ the night-futures return of that night and $h^{*}$ the ratio that makes the residual variance smallest. This regression is not estimated on the screen, so no particular value can be written down, and the ratio is a quantity to be estimated in a study that fixes a sample and a period, as Yeom, Baek, and Ryu (2013) did in examining the performance of strategies that use night futures.

The second use is the deviation of the night price from fair value. When the night price rises above the fair value the cost-of-carry relation sets, selling the futures and buying the replicating portfolio in the regular session is the direction, and when it is pushed below, the reverse holds (Cornell and French 1983). Cash and futures converge at expiry, so the gap reverts by construction, but the time reversion takes and the funding cost over that time decide the realized result. At night the cash basket cannot be handled alongside, which widens the no-arbitrage band relative to the day, and the relation of Nam et al. (2006), in which futures lead the cash market in the regular session, has no counterpart at night.

The third use is reading the shape of night volume and night movement themselves. Barclay and Hendershott (2003) report that after-hours volume is low while the share of the price move carried by a single trade is large, and in the stealth-trading result of Barclay and Warner (1993) mid-sized trades pull cumulative price change disproportionately. The two together explain why trades accumulating quietly in a thin book leave so large a mark. The seasonality of Webb et al. (2016), by which the price impact of a trade is not constant through the day, means a trade of the same size moves the price differently depending on the hour, so night volume carries its information in the comparison with the usual level in the same hours rather than in its absolute size.

The fourth use is marking overnight exposure in real time. Daily settlement on its own leaves the profit and loss of the night to surface only the next morning, whereas the night-futures price leaves a market price through that stretch as well. If the shape Bondarenko and Muravyev (2023) and Boyarchenko, Larsen, and Whelan (2020) observed in other markets, most of the daily return concentrated in a narrow band outside regular trading hours and around the European open, is present here too, then leaving the night blank in a valuation leaves a substantial part of the risk blank. Ahn et al. (2024), who show that retail flows track the gap between overnight and intraday returns, indicate that the gap is bound up with the participant mix as well, and the measurement of Seon and Joo (2019), in which regular-session frictions changed after night trading began, says that this use reaches past one investor's profit and loss to the cost borne by the market as a whole. The trades Lee and Woo (2013) worked from in measuring price-discovery efficiency in the night session are the same trades that make the bars on this screen.

This screen is not a sample pinned to a fixed date but a live series that passes a session along while it is still running. The platform's testing framework scores only snapshots taken at a fixed point in time and leaves one harness row per series stamped with the sample's end date, and the night-session bars have no such row. No battery of unit-root, stationarity or structural-break tests is therefore reported and no verdict is claimed. Not having run a test is not the same thing as having passed one.

The standing of the numbers on the screen follows from this. The open, the high, the low, the close, the volume and the change against the reference price are constructed quantities made from trades by a fixed rule, and they are not values a model estimated. A constructed quantity carries neither a standard error nor a null hypothesis, so the word significance has nothing to attach to, and a value that moves during a session moves because more trades have arrived rather than because an estimate was updated. The bar count and the presence of minutes without trades are likewise properties of the data itself and not products of missing-value handling.

The condition under which a battery becomes possible is plain. Once completed sessions are archived as a panel stacked by session date and that panel's end date is frozen, a list of tests suited to the character of the series can be fixed in advance, run once and recorded together with the direction of each verdict. Until that sample exists, this series has no test results.