What the number means
A futures-implied cash value for the session date shown. New trades and news can change it before the opening auction.
How implied opening prices are calculated from futures markets.
A futures-implied cash value for the session date shown. New trades and news can change it before the opening auction.
Read the session date, quote age and source together. A connected stream does not make an old underlying quote current.
Use the recorded accuracy and sample size. The volatility cone describes market movement, not the model’s historical prediction error.
Cost of carry relates the futures price to the current cash value through financing, dividends and time to expiry. Removing that carry gives an index-equivalent price implied by the future:
Rearranging to solve for the implied spot price given today's futures price :
Start with a future at 6,000 and a cash reference of 5,970. Assume a 4% annual financing rate, a 1.5% annual dividend yield and 30 calendar days until the quoted contract expires.
Rates enter as decimals. Carry accrues over calendar time; the 252-trading-day convention belongs to the volatility calculation.
Financing exceeds dividends, so the implied cash value is below the futures price. This adjustment is not the predicted market move.
Discount the future using the exponential formula above. Keep full precision until the displayed result is rounded.
The implied move compares fair value with the cash reference, not the futures contract’s own previous settlement.
Holding the future fixed, a higher financing rate lowers implied cash value; a higher dividend yield raises it. As expiry approaches, the carry adjustment shrinks.
Stock cards show regular-session or extended-hours trading prices, rather than a futures-based prediction. Pre-market normally runs from 04:00 to 09:30 ET, and post-market from 16:00 to 20:00 ET; exchange holidays and shortened sessions affect availability.
Extended-hours trading can have lower volume and wider spreads. The session label identifies the quote being shown, while the change is measured against the relevant close.
Commodity cards compare the front and back futures contracts. Each contract’s month and price is shown so you can see the curve directly.
Financing, storage, insurance and immediate supply all affect this spread. Storage costs are estimates, and expiry rules differ by commodity. A rising curve does not by itself predict a rising spot price.
/ES, /NQ, /YM, /RTY), falling back to Yahoo Finance. For international indices (FTSE, DAX, Hang Seng, Nikkei 225): scraped from Investing.com, with delayed exchange-quote fallbacks when the scrape fails (FTSE via CNBC's quote API, DAX via onvista). KOSPI 200 uses whichever of the KRX day future (Naver realtime) and the KRX overnight night-session future traded most recently. TAIEX likewise prices off the fresher of the TAIFEX day and after-hours (night) markets via the exchange MIS API. For commodities: front-month futures via Schwab / Yahoo Finance (/GC, /CL, etc.).The short-term risk-free rate of the currency each index is denominated in — because the cost of carry finances the underlying basket in its own currency, so the discount must use that currency's rate (not a single global one). Per currency, and why each rate was picked:
USD = Treasury.gov 3-month bill — kept because it is empirically near-perfect: decomposing every scored US session shows the carry discount cancelling the futures basis to within a few basis points.
GBP = Bank of England SONIA and EUR = €STR — both overnight benchmarks, on the principle that index-futures arbitrage is financed at collateralised/overnight rates, not at credit-sensitive term interbank rates. (EUR previously used 3-month Euribor, which sits ~25–30bp above €STR and measurably over-discounted the DAX; switched 2026-08-05.)
HKD = HKAB HIBOR and TWD = TAIBOR — daily term fixings; no free overnight-benchmark feed exists for either, and with the Hang Seng and TAIEX on monthly contracts, t is so short that the term-vs-overnight distinction is worth <2bp on the open.
JPY and KRW have no free daily feed at all (TIBOR is a daily PDF, the Korean CD rate is key-gated), so they use the OECD 3-month interbank series via FRED — monthly, with a one-to-two-month publication lag, but live and auto-updating rather than a frozen constant.
Each rate falls back to a recent constant if its fetch fails or goes stale, and rates are refreshed a few times a day. Using one US rate for every market — as earlier versions did — biased non-USD implied opens by up to tens of basis points and could even flip the predicted direction on longer-dated contracts.
For equity indices and stocks: the trailing 12-month dividend yield of the index's proxy ETF (e.g., SPY for S&P 500), sourced from Yahoo Finance and sanity-clamped against a curated per-index estimate. Total-return indices are an exception — the DAX reinvests dividends into the index itself, so its carry uses q = 0 (subtracting a dividend yield would double-count).
The Nikkei 225 is a second exception: Japanese companies pay in two concentrated bursts (fiscal-year-end finals go ex in late March, interims in late September), while the OSE contract expires the second Friday of those months — before the burst. A smooth annual q therefore credits the carry with dividends the contract never carries (we measured a persistent +0.6% over-prediction, of which ~+0.2% was this effect). The Nikkei instead uses an effective q: the annual yield × the fraction of the year's dividend points actually going ex before expiry — ≈0 for most of each quarter, spiking only in the weeks a contract spans a burst. The q shown in the drilldown is this effective value.
Commodity storage and insurance costs are entered as a negative q. They add to financing in r − q, so a positive storage cost lowers the cash value implied by a fixed futures price.
The expiry must belong to the contract whose price is quoted. A front-month symbol can roll to a new contract before the old one expires; pairing the new price with the old expiry can create an artificial jump.
| Instrument | Contract | Expiry Rule |
|---|---|---|
| S&P 500, Nasdaq, Dow, Russell | /ES /NQ /YM /RTY | Third Friday of Mar/Jun/Sep/Dec |
| Nikkei 225 | Investing.com | Second Friday of Mar/Jun/Sep/Dec |
| FTSE 100 | Investing.com | Third Friday of Mar/Jun/Sep/Dec |
| DAX | Investing.com | Third Friday of Mar/Jun/Sep/Dec |
| Hang Seng | Investing.com | Second-to-last business day of every month |
| KOSPI 200 | Naver realtime (KRX day future) | Second Thursday of Mar/Jun/Sep/Dec |
| TAIEX | TAIFEX | Third Wednesday of every month |
| Gold, Silver, Copper | /GC /SI /HG | Last business day on or before the 25th |
| Crude Oil, Natural Gas | /CL /NG | 3 business days before the 25th of prior month |
Data is cached server-side for 60 seconds and streamed to clients via Server-Sent Events. Quotes may be delayed by up to 15 minutes depending on the provider.
Before a cash session, the headline is the implied open for the displayed session date. During that session it is current fair value. After the close, the date advances to the next trading session using the venue calendar. Stocks show the appropriate regular or extended-hours quote instead of a carry-model prediction.
The solid chart is price history; supported futures overlays are rebased to the cash index and are not actual cash trades. The dashed cyan segment leads to the current fair-value estimate, not a time-path forecast. Shaded bands identify session gaps and lunch breaks; the volatility cone is an indicative daily one-sigma range, not a calibrated opening-error interval.
Stream connection, quote age and chart freshness are independent. A failed chart refresh retains its last validated bars with a warning and last-bar date. Market inputs identifies fixed fallback yields by tenor, and calculation details preserves source names and timestamps on every screen size.
Alongside each implied opening price, we display a one-standard-deviation daily range derived from that market's own 30-day implied-volatility index. Each index expresses annualised implied volatility as a percentage; converting to a daily dollar move:
where is the implied opening price computed by the cost-of-carry model, and the volatility index is expressed as a decimal (e.g. 18 → 0.18). The resulting range gives a rough sense of overnight uncertainty baked into options markets. Each card uses its own exchange's vol index rather than a single proxy — the US indices use the CBOE VIX, while international cards use their native equivalents: VKOSPI (KOSPI 200), TAIWAN VIX (TAIEX), the Nikkei 225 VI, VHSI (Hang Seng), VDAX-NEW (DAX) and VFTSE (FTSE 100). Where a native index is unavailable, the card falls back to the US VIX.
Each index is predicted in its own venue's pre-open window and scored against that session's actual open — so London, Tokyo, Seoul, Taipei, Hong Kong, Frankfurt and New York are each measured against their own open, not a single US clock. The miss is recorded as a percentage:
The headline is the absolute percentage miss for the date shown. RIGHT WAY / WRONG WAY describes whether the predicted opening direction was correct. Signed bias below shows whether the model tends to predict too high (positive) or too low (negative).
Mean Absolute Error over the trailing 30 calendar days. Direction-agnostic — measures average miss magnitude. Useful for calibrating how much confidence to place in the model's current projection.
Several markets open via a call auction whose first published print sits near the prior close until constituents trade. For those (S&P 500, Dow, Russell 2000, FTSE, DAX, Nikkei, Hang Seng and TAIEX) we score against the settled level ~15 minutes in rather than the stale opening tick. KOSPI 200 uses its official opening-auction print when available. Nasdaq-100 uses the initial open. TAIEX uses its settled intraday level; its official auction print is retained as a diagnostic.
They use different reference prices. A futures percentage change uses that contract’s previous settlement; the dashboard compares carry-adjusted fair value with the relevant cash reference.
The squares show direction calls, oldest to newest: green is right, red is wrong, and hollow means excluded. Their percentage scores only the included boxes. A correct direction can still have a large price miss; MAE measures that magnitude.
N counts scored sessions in the rolling calendar window, not elapsed days. THIN means fewer than seven scored sessions. Low-confidence sessions remain visible as excluded history and do not improve or worsen the headline score.
A result needs both a saved prediction and a usable opening reference. Closed markets, missing observations and excluded sessions can leave an older result on screen. Its date matters; an empty 30-day window is shown explicitly.
The carry model above treats dividends as a smooth annual yield . Real index futures on price-return venues discount discrete dividend points concentrated in ex-dividend seasons — Taiwan's Jul–Aug cluster puts the post-July-roll TAIEX contract roughly 2% below cash, Hong Kong's H-share season runs May–Aug, Japan clusters in Mar/Sep, and the UK has interim seasons worth ~10–25 bp — which a continuous yield structurally cannot express. To measure that gap we also run a second model: record the observed basis at each venue's cash close (same futures source, both prints required to be seconds-fresh so a frozen feed can never be recorded), then at the next pre-open convert the live future through it, . The observed basis absorbs the discrete dividend discount exactly and re-anchors itself on the new contract after every roll (on the roll day itself, and whenever no fresh anchor exists — first day, an aged observation beyond three sessions — the model falls back to the standard carry calculation).
Status: shadow. Anchored predictions are currently logged in the accuracy tracker alongside the headline carry predictions for FTSE 100, Nikkei 225, Hang Seng, KOSPI 200 and TAIEX — every displayed implied open and every headline accuracy number on this site remains the pure cost-of-carry model until a venue's shadow error record proves the anchor out. The DAX is exempt by construction (a total-return index reinvests dividends, so and the carry model is already exact), as are the US indices (real-time futures plus a smooth, diversified dividend stream keep the discrete-dividend error below measurement noise).