Macro ReviewHow to read
Outlook · Asset outlook · Options outlook

Options price a 12% chance of SPY falling 10% in three months

What do option prices imply for the next three months, and does the site's own model add anything to them?

Data to 17 Nov 2026 · updated 19 Nov 2026Quoted 16:00 New York, 17 Nov 2026, the vendor's end-of-day quotes
SPY: priced chance of a 10% fall in three months12%what the market prices · 17 Nov 2026
SPY: model's median path, three months+3.9%model alone +5.8% · 17 Nov 2026
Chance of finishing lower, three months28%of 4,000 model paths · 17 Nov 2026
Direction called, three months73%always-up 78% · 92 tests
175 of 176 assets covered on the latest session

In brief

What the options market prices for one asset, beside the site's own model and its record
  • Options price a 12% chance that SPY ends three months more than 10% lower.→ Exhibit 1
  • In SPY's forecasts the short-horizon band sits furthest from zero, at +0.07 standard units.→ Exhibit 3
  • The engine's forecasts for 16 Feb 2027 put SPY's median at +8.0%.→ Exhibit 4
  • The model's median path for SPY points +3.9% over three months.→ Exhibit 5
  • Its paths called SPY's three-month direction 73% of the time, against 78% for always-up.→ Exhibit 6
What would change this view
What would change this viewTriggerTodayDecided by
The model's paths beat always-upthree-month direction called more often than always-up73% against 78% (17 Nov 2026)not crossedeach run
Market and model point opposite waysthe priced median and the model's median on opposite sides of today's pricepriced +2.5%, model +3.9%not crossedeach run
The reasoning behind each point
  • Priced
    Options price a 12% chance that SPY ends three months more than 10% lower.

    What the market prices: the distribution of the price in 91 days that the session's option prices imply, 17 Nov 2026.

  • Components
    In SPY's forecasts the short-horizon band sits furthest from zero, at +0.07 standard units.

    The option-based forecasts at every horizon, summarised in three smoothed components.

  • Forecasts
    The engine's forecasts for 16 Feb 2027 put SPY's median at +8.0%.

    The engine's forecast fan on 16 Feb 2027: 7,675 paths drawn from its stored forecasts; their 10th to 90th percentile runs from −2.4% to +17.9%, their mean +8.0%.

  • Model
    The model's median path for SPY points +3.9% over three months.

    The site's model (model-conditional): the median of 4,000 paths given the engine's two-week path; +5.8% without it.

  • Record
    Its paths called SPY's three-month direction 73% of the time, against 78% for always-up.

    92 out-of-sample tests, re-fitted every month.

Next: Skew & term structure — which tail the same option prices favour

Source: marketdata.app; Macro Review calculations · data to 17 Nov 2026options.outlook-read

Options price a 12% chance of a 10% fall

Which way does the market lean over the next one and three months?

Exhibit 1 · Options price a 12% chance that SPY ends three months more than 10% lower.

SPY on the session of 17 Nov 2026

SPY is listed; it is drawn once the session folder ships its files.

About this exhibit
Price distribution options imply, one and three months ahead
Source: marketdata.app; Macro Review calculations · data to 17 Nov 2026options.outlook-1

Exhibit 2 · One-month risk reversal against its three-year range

SPY on the session of 17 Nov 2026

SPY is listed; it is drawn once the session folder ships its files.

Source: marketdata.app; Macro Review calculations · data to 17 Nov 2026options.outlook-2

The short-horizon band is doing the most work

Which shifts in the forecasts are doing the work now?

Exhibit 3 · In SPY's forecasts the short-horizon band sits furthest from zero, at +0.07 standard units.

SPY on the session of 17 Nov 2026

SPY is listed; it is drawn once the session folder ships its files.

About this exhibit
Step PCs: the forecasts' three horizon bands, raw and smoothed
Source: marketdata.app; Macro Review calculations · data to 17 Nov 2026options.outlook-3

The model's paths have not beaten always-up

Where do the engine's forecasts and the model's paths point, and how good have they been?

Exhibit 4 · The engine's forecasts for 16 Feb 2027 put SPY's median at +8.0%

SPY on the session of 17 Nov 2026

SPY is listed; it is drawn once the session folder ships its files.

About this exhibit
The engine's forecasts for each date: range, median and mean
Source: marketdata.app; Macro Review calculations · data to 17 Nov 2026options.outlook-7

Exhibit 5 · The model's median path for SPY points +3.9% over three months

SPY on the session of 17 Nov 2026

SPY is listed; it is drawn once the session folder ships its files.

About this exhibit
Daily price with the model's three-month range of paths
Source: marketdata.app; Macro Review calculations · data to 17 Nov 2026options.outlook-4

Exhibit 6 · Its paths called SPY's three-month direction 73% of the time

SPY on the session of 17 Nov 2026

SPY is listed; it is drawn once the session folder ships its files.

About this exhibit
The model's three-month record, out of sample
Source: marketdata.app; Macro Review calculations · data to 17 Nov 2026options.outlook-5

Past entries into today's regime moved a median +10.3% over 13 weeks

What happened after past readings like today's?

Exhibit 7 · Price paths after past entries into today's regime

SPY on the session of 17 Nov 2026

SPY is listed; it is drawn once the session folder ships its files.

Source: marketdata.app; Macro Review calculations · data to 17 Nov 2026options.outlook-6

Exhibit 8 · Price paths while the regime lasted and after it ended

SPY on the session of 17 Nov 2026

SPY is listed; it is drawn once the session folder ships its files.

Source: marketdata.app; Macro Review calculations · data to 17 Nov 2026options.outlook-9

Exhibit 9 · What followed days like today, by whether the regime ended

SPY on the session of 17 Nov 2026

SPY is listed; it is drawn once the session folder ships its files.

Source: marketdata.app; Macro Review calculations · data to 17 Nov 2026options.outlook-10

Every covered asset

How do the assets compare on the same measures?
Every covered asset on the same measures
AssetGroup10% fall priced, 3mModel, 3mModel aloneCalledAlways-upFitTests
Equity index12%+3.9%+5.8%73%78%−0.3192
Equity index19%+4.7%+7.2%71%76%−0.2291
Equity index15%−2.1%+2.9%48%60%−0.7092
Equity index11%+4.6%+5.9%68%73%−0.0392
Equity index7%+4.2%+4.7%67%73%−0.1192
Sectors–+2.1%+1.9%53%62%−0.2792
Sectors9%+0.8%+4.5%70%74%−0.2450
Sectors17%+5.1%+5.8%47%70%−0.2130
Sectors–+1.3%+1.9%62%75%−0.1391
Sectors–+2.8%+5.5%60%72%−0.0292
Sectors–+3.8%+8.6%68%73%−0.1292
Sectors–+1.3%+0.2%72%74%−0.0192
Sectors–+1.6%+1.9%62%60%−0.0281
Sectors8%−2.8%−2.4%59%67%0.0192
Sectors–+2.1%−0.9%62%66%−0.0491
Sectors–+1.5%+2.8%62%68%−0.2392
Industries and themes31%−3.5%+4.3%70%74%−0.1791
Industries and themes33%+1.8%+9.2%62%67%−0.3069
Industries and themes–+4.5%+8.7%64%67%−0.0891
Industries and themes–+2.7%+3.1%46%61%−0.0890
Industries and themes–+1.8%+2.5%59%72%−0.2588
Industries and themes27%+1.4%+3.9%51%56%−0.1585
Industries and themes–+2.4%+5.0%53%65%−0.1986
Industries and themes29%+3.4%+12.4%45%64%−0.1286

Appendix

How the distribution is drawn from option prices, how the paths are modelled, and what each check holds4,866 checks, all pass
About this page

What the options market prices for one asset over the next one and three months, beside the site's own model of where its price goes and that model's record. The distribution comes from the latest session's option prices; the paths are model-conditional, made from the option-based forecasts' components, with and without the engine's two-week path, day by day for three months. The record sits open beside the paths, because a range of paths is only worth what its record shows.

What the figures are

Model-conditional: the paths come from a daily model of the return on the three smoothed components of the option-based forecasts, its lag chosen by the Bayesian information criterion (at most 21 sessions), and 4,000 draws from its forecast distribution with a fixed seed, 63 sessions (three months) out. The conditional paths hold the components to the engine's path for ten sessions — its trend slope decaying at the rate measured from the slope's own history — and leave the return free; after that the model runs on its own dynamics.

The price distribution

The price distribution is what the market prices: per expiry, the forward from put-call parity and a smile in total variance against log-moneyness (the raw SVI form); at 30 and 91 days the smile is interpolated in total variance between the bracketing expiries and the density read from the call prices' curvature across strikes. Two penalties in the fit keep the density a density: the wings' slope bound and no butterfly arbitrage across the drawn strikes. A horizon is printed only when its density sums to one, stays non-negative and is centred on the forward. Its smile must also sit within half a vol point of the quoted at-the-money volatility, read at the forward from the six nearest quotes. A horizon the chain does not bracket is not extrapolated.

The closes

The closes are the engine's adjusted series: splits and distributions are carried back, so a return across an ex-date is a total return. The engine's forecasts, its component scores and its fan are made on the same series. The price distribution is read against the session's traded close, the price the options are struck in. The engine's own test for a change of adjustment basis runs on every asset's closes. The lead asset must pass it, an asset with a split-sized break is left out by name, and a smaller one is listed below with its date.

The closes, checked

The engine's own test finds a change of adjustment basis in the closes of 11 assets, each smaller than the 25% at which an asset is left out: DIA 21 Sep 2026 (0.02%); TLT 5 Oct 2026 (0.42%); IEF 2 Sep 2026 (0.34%); IEF 5 Oct 2026 (0.34%); TIP 5 Oct 2026 (0.15%); FXA 5 Oct 2026 (0.15%); CSCO 2 Sep 2026 (0.39%); DELL 21 Sep 2026 (0.14%); NOK 28 Sep 2026 (0.38%); GEV 28 Sep 2026 (0.06%); EQT 5 Oct 2026 (0.32%); BX 5 Oct 2026 (0.94%). The engine's test for a change of adjustment basis was not run on 22 assets, so their closes are not tested: FX_AUDCAD, FX_AUDCHF, FX_AUDJPY, FX_AUDX, FX_CADCHF, FX_CADJPY, FX_CADX, FX_CHFJPY, FX_CHFX, FX_EURAUD, FX_EURCAD, FX_EURCHF, FX_EURGBP, FX_EURJPY, FX_EURX, FX_GBPAUD, FX_GBPCAD, FX_GBPCHF, FX_GBPJPY, FX_GBPX, FX_GLDSLV, FX_JPYX. It does not apply to an asset composed as a ratio of two funds. For 49 assets some of the engine's stored forecasts start from a price that is no longer the close on their date. A distribution or a split since they were stored moved the closes and not the stored forecasts, so the engine's scores and fan carry the gap. Largest: FX_GLDSLV 2.58% (45 of 128,610 forecasts); OWL 2.27% (11,023 of 12,068 forecasts); PFE 1.46% (18,522 of 22,689 forecasts); FX_GBPCHF 1.31% (52,014 of 79,641 forecasts); FX_GBPJPY 1.30% (78,393 of 105,993 forecasts); FX_GBPAUD 1.17% (104,871 of 132,498 forecasts); FX_GBPCAD 1.12% (52,014 of 79,641 forecasts); FX_GBPX 0.97% (79,272 of 105,849 forecasts); ….

The shared date axis

The risk reversal, the PCA series, the engine's forecasts and the model's paths share one date axis: three years of sessions to the session, then the 63 weekdays of the paths ahead, so a date sits at the same place on each.

The PCA series

The PCA series are the engine's own, copied: each day it projects the option-based forecasts at every horizon onto two fixed sets of loadings. The step PCs (SPC_G1–G3) average the forecasts within three horizon bands, 5–25, 50–100 and 125–250 days. The orthogonal PCs (PC1–PC3) project the whole forecast curve onto level, slope and curvature. The card draws the raw daily score and the engine's smoothed score on either set, over three years.

The engine's forecasts

Model-conditional: the fan is the engine's own, copied and not recomputed. It is the fan the engine's own publishing step draws from the engine's files up to this session. The engine's close after this session was not yet settled, so the step was run on its files cut at the session. The engine takes each stored forecast with more than 21 days to run and draws 25 straight paths from it. Each path runs from the close on the forecast date to an end price shifted by a random error the size of its forecasts' past misses. On each date the fan gathers every path that crosses it and weights later forecasts more, with a half-life of 60 days. The card prints the mean, the median, the 40th–60th, 25th–75th and 10th–90th percentiles and the mean plus and minus two standard deviations, beside the adjusted close. The engine holds every price of its fan between zero and 20 times its latest close. The card runs three years back and, ahead, as far as the paths: three months. A fan whose past closes are not this session's closes is not drawn, and the card says so.

The risk reversal

The risk reversal is the engine's own 25-delta reading, put less call implied volatility in vol points: 30 days is read linearly between its 25- and 50-day points, daily, over three years, against its own 10th, 50th and 90th percentiles. Positive means puts cost more than calls.

The model

Model-conditional: the paths come from a daily model of the return on the three smoothed components of the option-based forecasts, its lag chosen by the Bayesian information criterion (at most 21 sessions), and 4,000 draws from its forecast distribution with a fixed seed, 63 sessions (three months) out. The conditional paths hold the components to the engine's path for ten sessions — its trend slope decaying at the rate measured from the slope's own history — and leave the return free; after that the model runs on its own dynamics.

The record

The record re-fits the model on the data to each origin, every 21 sessions from the 504th session (252nd for a shorter history), and scores the 63-session return against the historical average and against always-up.

The episodes

The episodes are the engine's. Its regime is the sign of each smoothed component, and an episode is a run of days in one regime, out of sample only. The card has two ways to follow the past entries into today's regime, beside the current episode from its own entry close. Extended tracking follows every entry for 65 sessions on the engine's closes, whether or not the regime lasted. Its band, median and mean are taken on each day across every entry with a close that day. Its dash-dot line is the median of any 65 sessions that start between the first entry and the last. In regime only follows each episode for as long as it lasted. Its band, median and mean need five episodes still running. On days with two to four the median and the mean continue, faded, without the band, so the later days rest on the longer episodes only. Each entry's path on the closes equals the engine's own path for as long as its episode ran. Where that cannot be shown for every past entry of an asset, its entries are not followed and the card says why. The lower chart divides each return by the volatility at that entry. That is the standard deviation of the 63 daily log returns before the entry day, annualised, so a calm and a turbulent start are read on one scale.

The two groups

The two groups are the same entries, split on each day by whether the regime was still running. An episode is in the first group for as long as it lasted and in the second from the session after. Each group's median and its 10th to 90th percentile are drawn on days when it holds five or more entries. Which group an entry would fall in was not known on the day, so the split is hindsight. It describes what the regime's end went with, and is not a rule to act on.

The fork

The table counts what followed sessions like today's, on every covered asset together. A session is like today's when the asset was in the same regime. It counts only when the sign of its three components is the engine's regime that day. The rows sort those sessions by the distance of the component nearest zero. A component's distance is its score divided by that score's own standard deviation over every session before it (126 or more). For each row the table gives the share of sessions on which the regime ended within 21 sessions. It also gives the mean return over those 21 sessions when the regime ended, when it did not and on all of them. Returns are in units of the asset's usual move: the daily volatility of the 63 sessions before the session × √21. The last 21 sessions of a regime still running are left out, because their outcome is not known yet. A share or a mean over fewer than 30 sessions is not printed. The sessions overlap and the assets move together, so the rows are counts of what happened, not independent tests and not a forecast.

Days and the session

Days to expiry are calendar days (τ = days ÷ 365). The site's realised volatility is √252 × the sample standard deviation of daily log closes, overnight returns included (the author's decision of 3 Oct 2026 (D-2026-10-03-19)); the options engine states its own as annualised by None on trading days, overnight return excluded. The quotes are the vendor's end-of-day set for the session, stamped 16:00 New York, 17 Nov 2026; the options engine's results for a session are read only when they are complete and were written after the close had settled (the author's decision of 4 Oct 2026 (D-2026-10-04-02)).

Whose quotes these are

Option data: marketdata.app. Every figure derived from it on this page is computed here.

Decisions this page follows

This page follows the author's decision of 4 Oct 2026 (D-2026-10-04-01) and the author's decision of 3 Oct 2026 (D-2026-10-03-21) and the author's decision of 4 Oct 2026 (D-2026-10-04-05) and the author's decision of 4 Oct 2026 (D-2026-10-04-06) and the author's decision of 4 Oct 2026 (D-2026-10-04-08) and the author's decision of 7 Oct 2026 (D-2026-10-07-SKEW1).

Before you read

Macro Review is a personal research site, still being built.

Nothing on it is investment advice, a recommendation or a solicitation.

Check every figure for yourself. Each names its publisher — confirm it there before you rely on it.

Figures can be revised, delayed or wrong.

Cones, fans, scenarios, fair values and “what followed” tables are conditional on the model each page names: they describe, they do not predict.

No warranty is given; use is at your own risk.