Three Futures Readings, One Gap, and a Cash Open That Ignored All Three

Three readings, two contracts, a thousand points apart

Overnight this desk published one futures number and labelled it single-sourced: CME Nikkei futures settling at 57,350, which against Thursday’s 61,077.90 cash close indicated a 6.10 per cent gap. We said a 6 per cent indicated move deserved better than one reader. It now has two more readers, and they do not agree with it.

• 57,350 — CME settlement, one Tokyo publisher at 23:00 UTC. Implied gap 3,727.90 points, 6.10 per cent.
• 58,125 — Chicago Nikkei futures, a second Tokyo publisher’s headline at 20:15 UTC. Implied gap 2,952.90 points, 4.83 per cent.
• 57,125 — SGX Nikkei 225 futures opening level, a third publisher at 23:32 UTC. Implied gap 3,952.90 points, 6.47 per cent.

The two Chicago readings are 775 points apart on what is nominally the same contract. We cannot tell you which is right and we are not going to pick. They carry different timestamps three hours apart, which is a sufficient explanation on a night when the Nasdaq fell 5.52 per cent, and it is also consistent with two publishers reading different contract months or different session marks. What we can say is that the band is 1,000 points wide and 1.64 percentage points wide, and that this is the second consecutive session on which this desk has had to publish a range where a feed would have given you a number.

Treat the spread as your error bar rather than as a finding about vendors. Any rule that would have behaved differently at 4.83 per cent than at 6.47 per cent was, last night, a coin flip dressed as a threshold.

The open was 1.15 per cent and it told you nothing

One publisher gives the Tokyo open at 60,374.35, down 703.55 points. That is the only reader we have for it and it is load-bearing here, so we are flagging it in the text rather than the footer. Against the 61,077.90 close it is a fall of 1.15 per cent.

Set that against the futures. The cash open printed 2,249.35 points above the highest of the three futures readings and 3,249.35 above the SGX opening level. The most pessimistic futures mark and the actual opening print disagreed by more than five per cent of the index. Whatever the futures were pricing, the auction that sets the cash open did not clear there.

And then the session went and got it. The intraday low is itself a band: 56,745 on one reader, described as about 4,332 points lower; 57,201.79 on a second, 3,876.11 lower; and “the 56,700s” on a third, which agrees with the first. Two of three readers put the low below 57,000, the band is 457 points wide, and the lowest reading sits below every one of the three futures marks. The morning session closed at 57,858.61, down 3,219.29 points or 5.27 per cent, with the Topix at 3,726.50, down 213.89 or 5.43 per cent. One publisher dates the break of 60,000 as the first since 20 May and another the level as the lowest since 14 April, the first claim intraday and the second not specified, so we are reporting both and asserting neither.

What a gap filter actually measures

This is the part that generalises past Japan. A gap rule is almost always written against the opening print, because the opening print is the one cheap, unambiguous observation available at the start of a session. The implicit model is that a dislocation announced overnight gets expressed in the auction and then the session trades normally around the new level.

Today falsified the model in the most inconvenient available direction. The dislocation was announced overnight, was not expressed in the auction, and was then expressed over the following ninety minutes as ordinary intraday trading. Every observation a gap filter takes — the open against the prior close, the first bar, the first five minutes — was consistent with a 1 per cent down day. The filter did not fail to detect a gap. It correctly detected that there was no gap, and the absence of the gap was the thing that mattered.

So the useful rewrite is not a tighter threshold. It is a different input. The observable that carried the information last night was the discrepancy between the futures marks and the cash close, not the opening print — and that discrepancy was available before the session opened, for free, in the form of a 1,000-point disagreement between two publishers. The operational version, for a system that cannot see order books: when the futures-to-cash spread is this wide, treat the first two hours of the cash session as the gap, not the first two minutes. Size as though the open has not happened yet, because on the evidence of today it has not.

That is a timing decision and a sizing decision, and it is deliberately not a directional one. We have no view on whether Japan is cheap at 57,000. We have a view on which minute of the day you are allowed to assume the repricing is done, and today the answer was two and a half hours after the bell.

Breadth inverted, which means the same trigger changed meaning overnight

Twelve hours ago our Asia desk published that Thursday’s Tokyo fall was a concentration event wearing an index’s clothes — the Nikkei down 1.79 per cent while the Topix rose 1.16 per cent, 1,127 Prime Market issues advancing against 408 declining, and one reader putting about 1,097 of the 1,110.73 index points on Advantest alone. The piece said a Topix-keyed or breadth-keyed rule would have correctly done nothing on Thursday, and that New York’s session was the broad one and would not mean-revert.

Today scores it, and the numbers are unusually clean. The Topix fell 5.43 per cent, against a 1.16 per cent rise on Thursday. Breadth across the Nikkei’s own constituents was 3 advancing, 221 declining and 1 unchanged at the morning close; on the Prime Market, one reader counts 1,515 issues down, about 98 per cent of the market; all 33 sector indices fell, with none higher. And the concentration arithmetic inverts: the four largest negative contributors together accounted for 963.48 of the 3,219.29 point fall, which is 29.9 per cent. On Thursday one name was most of the move. Today the top four were under a third of it.

The consequence for a system is specific and slightly unpleasant. A rule keyed on the Nikkei percentage was wrong on Thursday and right today, on the same trigger value, because the trigger does not carry the information. The information is in the second index and the advance-decline line, both of which were published in the same minute as the headline on both days, and neither of which is in most people’s feed. If you want one observable: Thursday the Topix and the Nikkei moved 2.95 percentage points apart in opposite directions, and today they moved together to within 16 basis points.

The correlated leg did not pay, and that is the day’s real risk lesson

The standing argument for watching Tokyo equities from an FX desk is that the yen crosses inherit the risk-off bid, so three yen positions on a bad Tokyo morning are one position wearing three costumes. Today the equity market delivered a five per cent rout and the FX leg delivered 75 pips.

Dollar–yen traded 157.25 to 158.00 on one reader’s range, 157.26 to about 158.00 on a second. The Tokyo fixing published by one bank was 157.77. Euro–yen fell to 176.47 and recovered to about 177.16, and sterling–yen was quoted around 212.45. One desk’s 03:18 UTC note describes risk-averse yen buying dominating on the equity collapse and then says plainly that the yen buying had already run its course; another, at 02:48, says the yen bid was unwound once the index decline stopped widening, with dollar buying ahead of payrolls supporting the pair.

Put the two sides next to each other. A 5.27 per cent move in the index you were using as your risk input produced a three-quarter-figure range in the pair you were sizing off it, and the pair finished the session nearer the top of that range than the bottom. If your exposure model converts an equity-volatility input into an FX position size, today it would have had you largest in the instrument that moved least. The correlation was not absent — the yen did bid, in the first hour — it was short-lived, and a correlation that decays inside one session is a worse input than no correlation, because it is right long enough to be trusted.

One more structural note, because it bears on where circuit-breaker risk actually lives. Korea’s composite index fell 8.76 per cent at worst and a 20-minute trading halt was triggered, lifted by about 00:36 UTC on one publisher’s headline. No page we read reports any Tokyo halt. A correlated Asian basket is not one jurisdiction’s microstructure: on the same shock, one venue stopped trading and the other did not, and a basket preset that assumes continuous quotes across the region had a 20-minute hole in one leg this morning.

The window you now have to size for

The US October employment report prints at 13:30 UTC, and the Canadian employment report prints in the same minute — two readers confirm both. The New York option cut is at 15:00. Our Calendar desk published the consensus eight hours ago and the figure to hold onto is not the 65,000 everybody has but the 29,000 that printed last month against an identical 90,000 unanimity, with a three-month average near 51,000 and a breakeven around 30,000 on one reader.

What today adds is that the session’s realised range has already been spent by an equity market, in the yen, before the release that is supposed to set it. The implied volatility this desk has been tracking moved from 6.23 per cent to the high seven per cent area on the one-week tenor — the tenor that covers this afternoon — and the Tokyo session then produced 75 pips. Those two facts point opposite ways and both are true. The honest reading is that the week has been repriced and the day has not, and the resolution arrives at 13:30 in a minute that also contains another country’s labour market.

And the close is not settled as we publish. One desk had the index 3,552 points lower at 05:51 UTC, nine minutes before the 06:00 bell. Two untimed ticker widgets on two pages show 3,692.10 and 3,724.04 lower, 31.94 points apart. We are not printing a close we cannot name a source for.

What this does not tell you

The opening level is single-sourced and the whole first half of this article turns on it. One publisher gives 60,374.35, down 703.55, and the arithmetic against Thursday’s close checks exactly, which is a consistency test and not a second reader. If that figure is wrong, the gap-arrived-late argument is wrong with it.

We cannot reconcile the two Chicago futures marks and we are not sure they describe the same thing. Different timestamps, possibly different contract months, possibly different session marks. We are also carrying the SGX opening level on one reader, and we have no second reader for the CME settlement either — we have three numbers from three publishers and no two of them agree.

No Tokyo trading halt is mentioned by any page we read, which is not the same as there having been none. Absence of a report is not confirmation, and we have no exchange statement. The Korean halt is reported by one publisher and its lifting by another.

The Thursday concentration figure we are scoring against was approximate in its own source and contradicted in magnitude by a second reader who put three chip names at about 1,800 points against an index that only fell 1,110.73. We are confident Thursday was concentrated. The 98.8 per cent is not a number to build on, and the 29.9 per cent for today is computed from one publisher’s contribution table.

And we have no yield curve. No page we read today carries a Japanese 10-year level beyond “below 3 per cent”, and the only US 10-year readings we have are a headline saying the yield fell on risk aversion with no level, and last night’s 4.90 per cent opening quote. Yesterday’s 9.26 basis point bull steepening in 2s30s was the cleanest evidence that this is a genuine risk-off rather than a sector rotation, and we cannot tell you whether it continued through Asia. That is the single largest hole in this note.

Related

• FX event calendar — payrolls and Canadian employment in the same minute at 13:30 UTC, with the New York cut at 15:00.
• Signals — why the futures-to-cash spread carried the information last night and the opening print did not.
• EA presets — correlated yen-cross exposure sized off an equity-volatility input that produced 75 pips.