Macro ReviewHow to read
Cross Asset · Fair value · Fair value board

Only the 10-year and credit have fair values: the 10-year sits 100bp above it

What is each macro asset worth, and how much should each model be trusted?

Every row carries its own dateUpdated 17 Nov 2026
10-year, move to fair value−101bp, fair 4.115% · par 5.12% · 16 Nov 2026
Gap to the −41.5%S&P 500 · the model's value 4,338 · , no target · 16 Nov 2026record unproven
−39bpspread 93bp vs default-risk 132bp · Sep 2026
Gold, model level from the price−18.7%level $289, price +23% above · no pull back to the level · 13 Nov 2026no target
Oil model's error, × random walk2.0052-week targets, 1729 weeks to 6 Nov 2026no reliable fair value

In brief

The for each macro asset, its confidence and what would change it; , not forecasts
  • S&P 500: the model's value at the 4,338 (−42%) · · no target.→ Exhibit 1
  • The 10-year sits 100bp above its fair value on the at today's (16 Nov 2026).→ Exhibit 3
  • Corporate spreads sit 39bp below the spread default risk predicts, the premium's 10th percentile (Sep 2026).→ Exhibit 5
  • Gold sits 23% above its model level, a level it has shown no tendency to return to; the model fails the board's rule.→ Exhibit 6
  • Oil has no reliable fair value: the inventory model's slope flips sign between halves and misses by 2.0 times a random walk.→ Exhibit 8
What would change this view
What would change this viewTriggerTodayDecided by
The Committee's path moves off the SEP paththe 2026 or 2027 median funds rate ≠ 4.1%4.1% · 4.1% (16 Sep 2026)pendingFOMC projections, 9 Dec 2026
Near-term growth leaves its bandthe Earnings fan's median outside −4.9 to 17.9% a year5.77% (16 Nov 2026)pendingBEA GDP and corporate profits, 25 Nov 2026
The premium turns positive > 0bp−39bp (Sep 2026)not crossedthe Board's file for Oct 2026 (no published date)
The 10-year reaches the SEP path's fair value10-year ≤ 4.115%5.12% (16 Nov 2026)not crossedeach Treasury close
The reasoning behind each point
  • Equities
    S&P 500: the model's value at the yield-gap premium 4,338 (−42%) · record unproven · no target.

    The two-stage model's value 4,338 against 7,417 (16 Nov 2026); drawing its estimated inputs puts it between 2,912 and 6,386, the spread of the inputs, not a probability. At the hold path's fair 10-year, the yield the 10-year row calls fair, the gap would be −29.7% instead of −41.5%. Record: Record unproven on this board's rule, on the Valuation history page's record of the gap (real-time data): the gap's 10-year slope is +0.038 per point with a Hodrick t of 1.61, short of the 1.96 the rule asks (4 non-overlapping windows); out of sample it loses to the historical mean at every horizon (R²_OS −0.05 at one year, 39 windows; −0.38 at 10 years, 2 windows); the record cannot tell it apart from the earnings yield less the real yield at any horizon (encompassing p 0.44–0.73; a test with little power on this few windows, so this is not evidence that it adds nothing). Its benchmark — the earnings yield less the history's real yield, the 10-year less trailing ten-year CPI inflation, not the TIPS yield — does no better on the same months (out-of-sample R² −0.18 at 10 years, 2 windows), though on the long record it scores +0.26 since 1926 (9 windows).

  • Rates
    The 10-year sits 100bp above its fair value on the SEP path at today's term premium (16 Nov 2026).

    Fair 4.115% on the SEP path, 4.065% on a hold, against a par yield of 5.12%. The SEP path at the grid's term premia, 2015–19 ACM average (−0.29pp) to today +50bp (+1.23pp); the five published paths at today's premium span 3.94–4.17%: 3.10–4.62%, an assumption range, not a probability.

  • Credit
    Corporate spreads sit 39bp below the spread default risk predicts, the premium's 10th percentile (Sep 2026).

    +39bp to fair value: the excess bond premium is −39bp (10th percentile since 1973). In sample (the Board re-estimates the split on the whole history with each update): as a 12-month target the fair spread missed by 75bp root-mean-square against 85bp for today's spread, a ratio of 0.89, over 633 months since 1973. After the 61 months with the premium at or below today's, it rose over the next 12 months in 87% (median +25bp).

  • Gold
    Gold sits 23% above its model level, a level it has shown no tendency to return to; the model fails the board's rule.

    Model level — fails the rule: the breakeven's slope is −1.84% per point (t = −1.08), the wrong sign; as a 12-month target the level misses by 1.34 times the random walk's error over 178 origins. The level ($289) is where the drivers' fitted changes, drift included, would have taken gold from the anchor — not a level the price returns to: the gap has not reverted (ADF p 0.71). Refitted from each January 2006–2016 it runs $233–$298 (an assumption range: the sample start); its range from past misses is $211–$403. Since Jan 2022 the model leaves +57% of the price's move unexplained (+0.99% a month against −0.30% before), and the full-sample drift (+0.84% a month, against +0.49% before Jan 2022) carries part of it into the level itself; the site has no source for the likely cause, official-sector buying.

  • Oil
    Oil has no reliable fair value: the inventory model's slope flips sign between halves and misses by 2.0 times a random walk.

    The full-sample slope is +0.95% of price per 1% of stocks above the five-year average with t = 1.69, short of the negative slope with |t| ≥ 2 the rule asks for; the slope changes sign between the halves (+1.43% to 20 Apr 2007, −1.56% from 27 Apr 2007); as a 52-week target it misses by 2.00 times the random walk's error over 1729 origins. Slope +0.95% of price per 1% of stocks (t = 1.69, R² 0.02); +1.43% (t = 0.9) to Apr 2007 and −1.56% (t = −8.3) since Apr 2007; as a 52-week target, error 2.00 times the random walk's over 1729 weeks. The dollar is not modelled: no source the site reads carries foreign interest rates, so a rate-differential model has no inputs.

Next: Valuation history — whether the gap has told us anything

Source: U.S. Treasury; Federal Reserve Board; New York Fed; Yahoo Finance; EIA; Robert J. Shiller data; FactSet; Macro Review calculationscross.fair_value-read

What is each asset worth against its price?

Each asset's model value and range as a percent of its price, the price at zero; then one line per asset, its inputs, method and record a click away

Exhibit 1 · Each asset's model value and range, as a percent of its price

Model value (dot) and range, % of the price, yield or spread, latest at zero; ranges labelled by kind; latest date for each.
About this exhibit
Range chart: each asset's model value (dot) and range as a % of its price, yield or spread (S&P 500, S&P 500 with the 10-year on hold, 10-year, credit, gold); the price at zero.
Each asset's model value against its price
Each asset's model value (dot) and range as a percent of its price or level, today at zero — for the 10-year and credit a percent of the yield or spread (the 10-year: fair 4.12% on the SEP path against 5.12%; credit: default-risk spread 132bp against 93bp). Crude oil (WTI): no reliable fair value; The dollar: not modelled.
Each bar is the asset's own method on this board, in its own unit before it is expressed against the price: the S&P 500's two-stage model (its input range), the 10-year's policy path plus term premium (assumed premia), the credit spread's default-risk part (its past misses) and gold's change model (its past misses). Model-conditional; no bar is a target.
Source: U.S. Treasury; Federal Reserve Board; Yahoo Finance; Robert J. Shiller data; Macro Review calculations · data to 16 Nov 2026cross.fair_value-14

Exhibit 2 · Each asset's model value and gap with its range and confidence

One row per asset: price, model value, gap (%), range and kind, and confidence, latest date for each; open a row for inputs, method and record.
AssetPriceThe model's valueGapRangeConfidence
S&P 500

Method. The two-stage identity (H-model): trailing earnings, near-term growth fading to the long-run rate, discounted at the 10-year plus the yield-gap premium

Inputs. Payout k · trailing earnings yield 0.845 · 3.985 (16 Nov 2026) · Near-term growth (the Earnings page's fan, bridged) 5.77% (16 Nov 2026) · Long-run growth (the SEP's longer-run real GDP + PCE) 4.0% (16 Nov 2026) · Fade, years (H) 6 · 3 · 10-year (rf) + yield-gap premium 5.12 · 5.17 (16 Nov 2026) · Cost of equity, at the model's premium 10.29% (16 Nov 2026)

At the hold path's fair 10-year. S&P 500, 10-year on hold: −29.7% instead of −41.5%.

Record. Record unproven on this board's rule, on the Valuation history page's record of the gap (real-time data): the gap's 10-year slope is +0.038 per point with a Hodrick t of 1.61, short of the 1.96 the rule asks (4 non-overlapping windows); out of sample it loses to the historical mean at every horizon (R²_OS −0.05 at one year, 39 windows; −0.38 at 10 years, 2 windows); the record cannot tell it apart from the earnings yield less the real yield at any horizon (encompassing p 0.44–0.73; a test with little power on this few windows, so this is not evidence that it adds nothing). Its benchmark — the earnings yield less the history's real yield, the 10-year less trailing ten-year CPI inflation, not the TIPS yield — does no better on the same months (out-of-sample R² −0.18 at 10 years, 2 windows), though on the long record it scores +0.26 since 1926 (9 windows).

Why no target. Record unproven: the gap's 10-year slope is +0.038 per point with a Hodrick t of 1.61, short of the 1.96 the rule asks (4 non-overlapping windows); out of sample it loses to the historical mean at every horizon (R²_OS −0.05 at one year, 39 windows; −0.38 at 10 years, 2 windows); the record cannot tell it apart from the earnings yield less the real yield at any horizon (encompassing p 0.44–0.73; a test with little power on this few windows, so this is not evidence that it adds nothing)

The gap's record →

7,41716 Nov 20264,338−41.5%2,912–6,386spread of inputsrecord unproven
10-year Treasury

Method. Policy path + term premium: the path's ten-year average short rate plus today's ACM term premium (the policy-rules grid)

Inputs. One further +25bp (the SEP path): ten-year average short rate 3.390% · Hold at IORB: ten-year average short rate 3.340% · ACM 10-year term premium 0.725pp (13 Nov 2026) · 10-year par yield 5.12% (16 Nov 2026) · Committee's median funds rate, 2026 · 2027 4.1 · 4.1 (16 Sept 2026)

Record. None out of sample: an identity — a path's average short rate plus a premium — so what can be wrong is an input. The ACM model that supplies the premium fits today's par 10-year to 0.9bp (5.111% fitted, 13 Nov 2026).

5.12%16 Nov 20264.115%−101bp3.10%–4.62%assumed valuesan identity: no out-of-sample record
Corporate credit (GZ spread)

Method. Gilchrist–Zakrajšek: the spread splits into the part issuers' expected default predicts (fair) and the excess bond premium (the gap)

Inputs. GZ credit spread 92.59bp (Sep 2026) · Predicted by default risk 131.78bp (Sep 2026) · Excess bond premium -39.19bp (Sep 2026) · Premium's percentiles since 1973: 10th, 25th, 50th, 75th, 90th −38/−25/−5/+21/+69bp (Sep 2026)

Record. In sample (the Board re-estimates the split on the whole history with each update): as a 12-month target the fair spread missed by 75bp root-mean-square against 85bp for today's spread, a ratio of 0.89, over 633 months since 1973. After the 61 months with the premium at or below today's, it rose over the next 12 months in 87% (median +25bp).

93bpSep 2026132bp+39bp94bp–200bppast missesan in-sample split
Gold (GLD)

Method. Change model: log GLD on the 10-year real yield, the log broad dollar (H.10) and the 10-year breakeven, in monthly changes, Feb 2006 – Oct 2026, Newey–West (6 lags); the level is the fitted changes accumulated from Jan 2006, anchored to the price's average over the sample

Inputs. GLD close $355.45 (13 Nov 2026) · 10-year real yield (TIPS par) 2.91% (13 Nov 2026) · Broad dollar index (H.10) 122.65 (13 Nov 2026) · 10-year breakeven (nominal − real par) 2.24% (13 Nov 2026) · Slope on the real yield, % per point (t) -7.53 · -4.64 · Slope on the dollar, % per 1% (t) -1.04 · -4.62 · Slope on the breakeven, % per point (t) -1.84 · -1.08 · Drift, % a month (t) 0.84 · 3.68

Record. Model level — fails the rule. In monthly changes, Feb 2006 – Oct 2026 (249 months, R² 0.31): real yield −7.53% per point (t = −4.6); dollar −1.04% per 1% (t = −4.6); breakeven −1.84% per point (t = −1.1); drift +0.84% a month. As a 12-month target the level's error is 1.34 times the random walk's over 178 origins, and the gap between the price and the level has shown no tendency to close (ADF p 0.71). Since Jan 2022 the model leaves +57% unexplained.

Why no target. Model level — fails the rule: the breakeven's slope is −1.84% per point (t = −1.08), the wrong sign; as a 12-month target the level misses by 1.34 times the random walk's error over 178 origins

Gold's model in full →

$355.4513 Nov 2026$288.81+23.1% above the level$211.38–$403.45past missesmodel level — fails the rule
Crude oil (WTI)

Method. Inventory model: log WTI spot on commercial crude stocks against their five-year average, weekly, 14 Aug 1987 – 6 Nov 2026, Newey–West (52 lags)

Inputs. Commercial crude stocks (excluding the SPR) 414.4mb (6 Nov 2026) · Five-year average for the week (range 427.6–439.4mb) 433.0mb (6 Nov 2026) · Stocks against the five-year average −4.3% (6 Nov 2026) · Slope, % of price per 1% of stocks (t, 2040 weeks) 0.95 · 1.69

Record. Slope +0.95% of price per 1% of stocks (t = 1.69, R² 0.02); +1.43% (t = 0.9) to Apr 2007 and −1.56% (t = −8.3) since Apr 2007; as a 52-week target, error 2.00 times the random walk's over 1729 weeks.

Why no number. the full-sample slope is +0.95% of price per 1% of stocks above the five-year average with t = 1.69, short of the negative slope with |t| ≥ 2 the rule asks for; the slope changes sign between the halves (+1.43% to 20 Apr 2007, −1.56% from 27 Apr 2007); as a 52-week target it misses by 2.00 times the random walk's error over 1729 origins

414.4mbcommercial crude stocks, not a price (the model's input) · 6 Nov 2026no reliable fair value––no reliable fair value
The dollar

Method. Rate differentials

Why no number. No source the site reads carries foreign interest rates, so a rate-differential model has no inputs.

–not modelled––not modelled
About this exhibit
Table: for each asset, price, model value, gap, range and confidence, with its method, inputs and record in an expanding row.
The board: each asset's model value, gap and confidence
One method per asset. Open a row for its inputs, method and record; every range says what kind of range it is, and no model here carries a target.
Source: U.S. Treasury; Federal Reserve Board; Yahoo Finance; U.S. Energy Information Administration; Macro Review calculations · data to 16 Nov 2026cross.fair_value-8

Each method, with its inputs

The 10-year against its fair values, the S&P 500 model's inputs, the spread against its default-risk value, and the two estimated models against the data

Exhibit 3 · The 10-year yield against its fair value on each policy path

10-year yield, %, daily since 16 Nov 2023; dashed, its fair values on the Fed's path (4.115%) and a hold (4.065%) at the latest term premium.
About
About this exhibit
Daily line: the 10-year par Treasury yield, with dashed fair values on the Fed's projected path and on a hold at the latest term premium.
The 10-year against its fair values
The Treasury-curve page's par 10-year since 16 Nov 2023, with the fair value of the SEP path (4.115%) and of a hold (4.065%) at today's term premium. The fair values are today's; the grid publishes no history of them.
Source: Copied from the Treasury-curve page (U.S. Treasury par yields) and What is priced (the policy-rules grid's fair values).cross.fair_value-1

Exhibit 4 · Inputs to the two-stage model for the S&P 500

Each input's value, %, and date, 16 Nov 2026; the range is the Valuation page's input band.
Range: spread of inputs
About this exhibit
Table of the two-stage model's inputs for the S&P 500, with value and date.
The S&P 500: the two-stage model's inputs
The two-stage identity (H-model): trailing earnings, near-term growth fading to the long-run rate, discounted at the 10-year plus the yield-gap premium. The identity's level is the index level × (1 + gap); its record on the gap's published history does not pass the board's rule, so no target. Range: the Valuation page's input band: the two-stage identity over random draws of its estimated inputs (near-term growth from the Earnings fan, the premium from its pool, each drawn independently of the other), 10th, 90th percentiles (gap −60.7 to −13.9%) — the spread of those inputs, not a probability for the index The identity's corner band, for comparison: 2,716–7,294 (assumption range).
Source: U.S. Treasury; Robert J. Shiller data via multpl.com; FactSet; Macro Review calculations · data to 16 Nov 2026cross.fair_value-9

Exhibit 5 · Corporate bond spreads against the spread default risk predicts, since 1973

The Gilchrist–Zakrajšek spread and its predicted part, basis points, since 1973; the distance between them is the excess bond premium, −39bp in Sep 2026.
About
About this exhibit
Two lines, monthly since 1973: the Gilchrist–Zakrajšek corporate spread and the part predicted by default risk; the distance is the excess bond premium.
Corporate spreads against the spread default risk predicts
The Gilchrist–Zakrajšek spread and its predicted part since 1973; the distance between them is the excess bond premium, -39bp in 2026-09.
Source: Board of Governors of the Federal Reserve System, the Gilchrist–Zakrajšek file; since 2000 copied from the Corporate spreads page.cross.fair_value-2

Exhibit 6 · Gold's price against its change model's level since 2006

GLD month-end close, $; dashed, where the model's fitted changes would have taken it; thin lines, the 10th and 90th percentiles of past gaps; since 2006.
About this exhibit
Monthly lines since 2006: GLD's month-end close, the change model's level (dashed), and the percentile range of past gaps. A model level, not a fair value.
Gold against its change model's level
GLD's month-end close against where the drivers' fitted monthly changes (real yield, dollar, breakeven, and a drift) would have taken it from the anchor, the price's average since Jan 2006. The level is not a price gold returns to: the gap has not reverted. The thin lines are where the gap has wandered (10th–90th percentile of its past values). The model fails the rule, so the line is a model level, not a fair value, and there is no target.
Source: Computed here from Yahoo Finance (GLD), U.S. Treasury par yields and the Board's H.10 broad dollar index.cross.fair_value-3

Exhibit 7 · US commercial crude stocks against their five-year range

Million barrels, weekly: stocks excluding the SPR against the same week's previous five years (average, low, high); 414.4mb, week to 6 Nov 2026.
About this exhibit
Weekly lines: US commercial crude stocks excluding the SPR against the five-year average, low and high for the same week (ten years by default).
Commercial crude stocks against their five-year range
The oil model's input: stocks excluding the SPR against the same week of the five previous years (average, low and high); 414.4mb in the week to 6 Nov 2026.
Source: U.S. Energy Information Administration, weekly stocks; five-year range computed here.cross.fair_value-4

Exhibit 8 · WTI's residual from the inventory model since 1987

Log WTI less its fitted value on crude stocks' gap from the five-year average, %, weekly since 1987; the spot price itself is not shown.
About this exhibit
Weekly line since 1987: the inventory model's residual, log WTI less its fitted value on stocks, %.
WTI against the inventory model
The model's residual: log WTI less its fitted value on the stocks' distance from the five-year average, in percent. The spot price's level is not displayed; the residual is what the model leaves unexplained. The model fails the stability rule, so the board prints no fair value.
Source: Computed here from EIA's weekly stocks and its daily WTI spot series (attributed by EIA to Refinitiv/LSEG).cross.fair_value-5

How far can each method be trusted?

Each method's record, the stability of the estimated slopes, the test that decides what may be printed, and what gold's model cannot explain

Exhibit 9 · Each model's track record against the test for a fair value

Full sample and each half: slope, t-statistic; target error ÷ random walk's (ratio); pass or fail on sign, stability and out-of-sample; the latest verdict.
About
Model · slopeSlope (t)First halfSecond halfTarget error × random walkSignHalves · addsOut of sampleVerdict
S&P 500the gap as a 10-year return signal · 4 windows · May 1975 – Nov 2016+0.038Hodrick t 1.61 (needs 1.96)adds to E/P − real yield: noencompassing p by horizon: 1y 0.59 · 3y 0.70 · 5y 0.73 · 10y 0.44R²OS −0.38against the historical mean · 2 windows; one year −0.05failsfailsfailsrecord unproven
Gold (GLD)real yield, % per point · 249 months · Feb 2006 – Oct 2026−7.53t -4.64 · R² 0.31−9.58t -5.5 · to May 2016−5.36t -2.6 · from Jun 20161.34178 origins · 12 aheadpassespassesfailsmodel level — fails the rule
dollar, % per 1%−1.04t -4.62−1.00t -3.0 · to May 2016−1.13t -4.3 · from Jun 2016passespasses
breakeven, % per point−1.84t -1.08−1.89t -0.9 · to May 2016−3.05t -1.2 · from Jun 2016failspasses
Crude oil (WTI)2040 obs · 14 Aug 1987 – 6 Nov 2026+0.95t 1.69 · R² 0.02+1.43t 0.9 · to 20 Apr 2007−1.56t -8.3 · from 27 Apr 20072.001729 origins · 52 aheadfailsfailsfailsno reliable fair value
  • 10-year Treasury

    None out of sample: an identity — a path's average short rate plus a premium — so what can be wrong is an input. The ACM model that supplies the premium fits today's par 10-year to 0.9bp (5.111% fitted, 13 Nov 2026).

  • Corporate credit (GZ spread)

    In sample (the Board re-estimates the split on the whole history with each update): as a 12-month target the fair spread missed by 75bp root-mean-square against 85bp for today's spread, a ratio of 0.89, over 633 months since 1973. After the 61 months with the premium at or below today's, it rose over the next 12 months in 87% (median +25bp).

  • The dollar

    No source the site reads carries foreign interest rates, so a rate-differential model has no inputs.

About this exhibit
Table: each model's slope with t-statistic, the two halves, target error against a random walk, the sign, halves and out-of-sample tests, and the verdict.
Each method's record, and the rule that decides what may be printed
The stability rule, fixed in the spec before the data are read: the full-sample slope has the sign theory gives it with |t| ≥ 2, the two halves share its sign, and the re-estimated fair value beats the random walk as a target (error ratio below 1). Oil's model prints “no reliable fair value” and no number if any leg fails. Gold's model is held to the same legs for each of its three slopes and graded: A, B or C, as its row says. The S&P 500's identity, once exempt as having no record, is graded the same way on the record the Valuation history page publishes for its gap: the sign and |t| (at that page's own threshold, 1.96, where gold and oil use 2), beating the historical mean out of sample, and adding to the earnings yield less the real yield.
Source: U.S. Treasury; Federal Reserve Board; New York Fed; Yahoo Finance; EIA; Robert J. Shiller data; FactSet; Macro Review calculationscross.fair_value-10

Exhibit 10 · Gold's rolling sensitivity to the real yield, dollar and breakeven

% change in gold per point of the real yield and breakeven, per 1% of the dollar; rolling 60-month windows since Jan 2011.
About this exhibit
Three lines: gold's rolling slope on the real yield, the dollar and the breakeven, % per point.
Gold's three slopes, 60-month windows
Each point is the change regression over the 60 months to that month. The theory's sign held in 100% of windows for the real yield (negative), 100% for the dollar (negative) and 13% for the breakeven (positive).
Source: Computed here from Yahoo Finance (GLD), U.S. Treasury par yields and the Board's H.10 broad dollar index.cross.fair_value-6

Exhibit 11 · Gold's price move the model cannot explain, summed over time

Monthly residuals of gold's change model, summed from Jan 2022, %.
About this exhibit
One line, monthly: gold change model's residuals summed from the drift's start month, %.
What the model cannot explain: the residual summed since Jan 2022
The monthly residuals of the change model, added up from Jan 2022: +57% by Oct 2026, an average of +0.99% a month against −0.30% before. The site reads no measure of official-sector buying, the likely cause; the model reports the drift and does not absorb it.
Source: Computed here from Yahoo Finance (GLD), U.S. Treasury par yields and the Board's H.10 broad dollar index.cross.fair_value-12

Exhibit 12 · Oil's rolling sensitivity to crude stocks, 260-week windows

Each point is the slope over the 260 weeks to that week; the theory's sign (negative) held in 78% of windows, since 11 Sep 1992.
About this exhibit
One line: WTI's rolling 260-week slope on crude stocks, % per %.
Oil's slope on stocks, 260-week windows
Each point is the slope over the 260 weeks to that week; the theory's sign (negative) held in 78% of windows.
Source: Computed here from EIA's weekly stocks and its daily WTI spot series.cross.fair_value-7

Where the inputs come from

The pages that own the models this board copies

Gold's model in full

The change regression on each half of the sample, the residual before and after 2022, and the cointegration test on the levels

Exhibit 13 · Gold's model by sample half, its long-run test and its refitted level

Slopes by half, % of gold per point of the real yield and breakeven, per 1% of the dollar; trace tests; level refitted each January 2006–2016, $.
SampleMonthsreal yielddollarbreakevenDrift
First halfFeb 2006 – May 2016124−9.58t -5.5−1.00t -3.0−1.89t -0.9+0.53t 2.0
Second halfJun 2016 – Oct 2026125−5.36t -2.6−1.13t -4.3−3.05t -1.2+1.09t 3.2
Whole sample249−7.53t -4.6−1.04t -4.6−1.84t -1.1+0.84t 3.7

What the model leaves unexplained: the residual averaged −0.30% a month over the 191 months before Jan 2022 and +0.99% a month over the 58 months since — +57% in all since Jan 2022. The site reads no measure of official-sector gold buying, the likely cause, so the model cannot say whether it is that.

Levels testedMonthsTrace, r = 05% criticalTrace, r ≤ 15% criticalReading
log gold, real yield, log dollar, breakeven24856.347.922.329.8rejects no cointegration at 5% (one relation)
log gold, real yield, log dollar24818.529.83.015.5no cointegration at 5%
real yield, log dollar, breakeven24839.729.89.815.5rejects no cointegration at 5% (one relation)

Where the one relation comes from: the breakeven on its own is stationary (ADF p 0.002, 1 lag); the restriction “the relation is the breakeven alone” is not rejected (LR 4.62, χ²(3) p 0.20). The narrower restriction “gold is not in it” is rejected at the 5% level (LR 4.00, χ²(1) p 0.045), so these tests do not settle whether the relation includes gold. The price's gap to the level has not reverted (ADF p 0.71). The drift is +0.84% a month on the whole sample and +0.49% on the 191 months before Jan 2022: the constant carries part of the later drift into the level.

Sample fromJan 2006Jan 2007Jan 2008Jan 2009Jan 2010Jan 2011Jan 2012Jan 2013Jan 2014Jan 2015Jan 2016
Level todaythe same model, the same anchor rule$289$278$266$262$259$251$233$242$278$280$298
Price above it+23%+28%+34%+36%+37%+42%+53%+47%+28%+27%+19%

The level refitted from each declared sample start: 233–298 dollars, assumed values — declared values of the start month, not a probability.

About this exhibit
Three tables: gold's change regression on each half and the whole sample; trace tests on the levels; the model level refitted from each sample start.
Gold's model in full: the halves, what it leaves unexplained, and the cointegration test
The change regression on each half of the sample (slopes in % of gold per point of the real yield and the breakeven, per 1% of the dollar; the drift in % a month), the residual's average before and after Jan 2022 with its sum since, Johansen's trace test on the levels, and the level refitted from each declared sample start. Gold, the real yield and the dollar share no long-run level relation. The one relation the test finds once the breakeven is added cannot be told from the breakeven alone, and the drivers without gold show it too. One test reads the other way: that gold can be dropped from it is rejected at the 5% level (the lines under the table). The model does not rely on a long-run relation. The level is not an attractor either: the price's gap to it has not reverted.
Source: Computed here from Yahoo Finance (GLD), U.S. Treasury par yields and the Board's H.10 broad dollar index.cross.fair_value-13

Appendix

Each method's specification, the , the ranges' kinds, the sources and the checks109 checks, all pass
About this page

Model-conditional values, not forecasts, targets or advice: what each macro asset is worth on stated inputs, one method per asset, each with its inputs, its range and what kind of range it is, and its record in words — a record too short to judge is unproven, and where a model fails its stability test or no model exists the board says so instead of printing a number.

Exhibit 14 · Each method's specification

  • Model-conditional

    Model-conditional: every fair value on this page is what a stated model implies on stated inputs — a description of the price against the site's models, not a forecast of where the price will go and not advice. One method per asset, printed with its inputs, its range and the range's kind, and its record; an asset whose model fails its stability test, or that has no model, says so; gold's model level and the S&P 500's identity level are printed with a grade that says how far their record lets them be read.

  • 10-year

    The 10-year: the fair value of a policy path is the path's ten-year average short rate plus the ACM term premium of the term-premium page's latest reading (0.725pp, 13 Nov 2026) — the policy-rules grid's cell for that path at today's premium, which What is priced publishes as the path's fair value. The SEP path (one further +25bp, then a hold through 2028 Q4, then the longer-run rate) is the headline; the hold path prints beside it. The range is the SEP path at the grid's other term premia — declared settings, so an assumption range, not a probability; the five published paths at today's premium are a scenario range. It is an identity, so its record is the fit of the model that supplies the premium.

  • S&P 500

    The S&P 500: the Valuation page's two-stage identity — price = k × trailing earnings × [(1 + g_L) + H(g_S − g_L)] ÷ (r − g_L) — with near-term growth 5.77% fading to 4.0% over 6 years and r the 10-year plus the yield-gap premium (10.29%). The gap is copied, and the range is the Valuation page's input band — the identity over independent random draws of its estimated inputs, not a probability for the index — copied and tied to this board's level; the level is the forward P/E × forward EPS the Scenarios page copies as the identity's level, and the fair value is the level × (1 + gap). The Valuation page's gap distribution — median −17.0% four quarters ahead — is the one-stage identity's joint draws (its own reconciliation row says so), so it is not this row's range. The row is graded on the gap's published record (Valuation history), as gold is on its own: the gap as a predictor of forward real returns at 10 years, held to the sign and a Hodrick |t| of at least 1.96, an out-of-sample R² above zero against the historical mean, and adding to the earnings yield less the real yield — the history page's own declared thresholds. Grade C: the level is printed as the identity's, with no target. The record's backtest error — the 10th to 90th percentile of the gap's out-of-sample misses — prints beside the level's own range; it is a backtest range, not a probability.

  • Credit

    Corporate credit: the Gilchrist–Zakrajšek spread is the average spread of senior unsecured corporate bonds over Treasuries of the same duration, across ratings; the Board splits it into the part a model of each issuer's expected default predicts and the excess bond premium, the part it does not. The predicted part is the fair spread and the premium the gap. The Corporate spreads page publishes the three series from 2000; the history back to 1973 is taken from the Board's file (read on 31 Oct 2026) and tied to the copy month by month. The range is the premium's own history — the default-risk model's past misses — a residual range. The investment-grade curve (HQM) is not decomposed, so the all-ratings spread stands in for investment grade.

  • Gold

    Gold's level on this board is model-conditional: it is where the drivers' fitted monthly changes, drift included, would have taken gold from the anchor — not a forecast, and not a level the price is drawn back to. The model was declared after an exploratory fit of two of its three drivers (the real yield and the dollar, in the scoping memo of 25 September 2026, which also chose the drift date) and before this estimate; the breakeven was the one driver not seen first. It regresses the monthly change in log GLD (month-end adjusted close) on the changes in the 10-year real par yield, the log broad dollar (the Board's H.10 index) and the 10-year breakeven (nominal less real par), Feb 2006 – Oct 2026, 249 months, Newey–West with 6 lags. Changes, because the levels share no stable long-run relation: log gold, real yield, log dollar: trace 18.5 against a 5% critical value of 29.8 (does not reject no cointegration). With the breakeven among the levels the trace test rejects (log gold, real yield, log dollar, breakeven: trace 56.3 against a 5% critical value of 47.9 (rejects no cointegration)). Most of the evidence puts that down to the breakeven's own mean reversion: the breakeven alone is stationary (ADF p 0.002); the one relation cannot be told from the breakeven alone (LR p 0.20); and the drivers without gold show the same relation (real yield, log dollar, breakeven: trace 39.7 against a 5% critical value of 29.8 (rejects no cointegration)). One test reads the other way on the sample to Oct 2026: that gold can be dropped from the relation is rejected at the 5% level (LR p 0.045). These tests therefore do not settle whether a long-run level relation includes gold, and the model does not rely on one. The level accumulates the fitted changes from Jan 2006, anchored so the price sits on it on average over the sample, and is carried to 13 Nov 2026 at that day's drivers. It has no pull: the price's gap to it has not reverted (ADF p 0.71), and as a 12-month target it loses to the random walk. It also moves with the sample start: refitted from each January 2006–2016 it runs $233–$298 (an assumption range). Slopes: real yield −7.53% per point (t −4.6), dollar −1.04% per 1% (t −4.6), breakeven −1.84% per point (t −1.1); drift +0.84% a month (+0.49% on the changes before Jan 2022, so the full-sample drift carries part of the post-2022 drift into the level). The grade is the stability rule's, driver by driver (|t| ≥ 2): A if every leg passes (a fair value; a target may follow), B if sign and |t| pass but the halves or the 12-month target test fail (an indicative fair value, no target), C if a sign or |t| fails (the model's level, labelled as failing, no target). Today: grade C. The residual range is where the gap has wandered — the 10th to 90th percentile of the price's distance from the level over the sample (−31% to +33%, log points) — not a band the price returns inside. What the model cannot explain: since Jan 2022 the residual has averaged +0.99% a month against −0.30% before, +57% in all; the likely cause, buying by central banks and other official holders, is not measured by any source this site reads (the World Gold Council's demand data and the IMF's reserve data are not used). No dummy and no break date absorb it, though the drift absorbs part of it. The Valuation page's 156-week beta of gold on the real yield, −4.3% per point, is the same sensitivity on weekly data.

  • Oil

    Oil: log WTI spot (the last price on or before each week's Friday, if no more than 7 days old — 2 weeks without one are left out; 1 non-positive day dropped) regressed on commercial crude stocks against the average of the same week in the five previous years (ISO weeks, week 53 read as 52), 14 Aug 1987 – 6 Nov 2026, 2040 weeks, Newey–West with 52 lags. The five-year average, low and high printed for the week are the Oil & petroleum page's, copied; the model's own band, which it needs back to 14 Aug 1987, equals that page's in every week both carry (a check). The rolling slope is over 260 weeks; the target test starts after 260 weeks and scores the fair value as the log price 52 weeks later against the random walk. The spot price's display rights are unresolved, so the page draws the model's residual and the stocks, never the price level.

  • The rule

    The stability rule, fixed in the spec before the data are read: the full-sample slope has the sign theory gives it with |t| ≥ 2, the two halves share its sign, and the re-estimated fair value beats the random walk as a target (error ratio below 1). Oil's model prints “no reliable fair value” and no number if any leg fails. Gold's model is held to the same legs for each of its three slopes and graded: A, B or C, as its row says. The S&P 500's identity, once exempt as having no record, is graded the same way on the record the Valuation history page publishes for its gap: the sign and |t| (at that page's own threshold, 1.96, where gold and oil use 2), beating the historical mean out of sample, and adding to the earnings yield less the real yield.

  • Dollar

    The dollar: not modelled. No source the site reads carries foreign interest rates, so a rate-differential model has no inputs. The board says so rather than print a level from a model with missing inputs.

  • Ranges

    Every range on the board carries its kind: an assumption range is declared values of an assumed input, a residual range the model's past misses, an input range the spread of estimated inputs carried through the model, a scenario range named paths. Assumption and scenario ranges are not probabilities.

  • Copies

    The 10-year, S&P 500 and credit rows are copies: each value is read from the owning page's published data through the site's manifest, with its path and date, and re-read from disk at each update, which is what the checks compare. Gold and oil are estimated here from the publishers' files; the dates on the falsifier rows come from the publishers' release calendars.

  • Licence

    Yahoo Finance prices, with attribution; Treasury, Board and EIA data are free to use; the WTI spot series is attributed by EIA to Refinitiv/LSEG, and only the model's residuals and slopes are printed from it.

Source: U.S. Treasury; Federal Reserve Board; New York Fed; Yahoo Finance; EIA; Robert J. Shiller data; FactSet; Macro Review calculationscross.fair_value-11

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.