Macro ReviewHow to read
Cross Asset · Return analysis · Sources of return

Over the last year the S&P 500 returned +11.3%, with earnings growth adding +20.7pp.

What is each asset's return made of: income, growth or a change in what the market pays for it?

Prices and curves to 16 Nov 2026Factor model copied from Factor analysis (16 Nov 2026)Updated 17 Nov 2026
S&P 500, 12 months+11.3%total return, dividends included · 252 sessions to 16 Nov 2026
S&P 500: income, 12 months+1.1ppdividends · to 16 Nov 2026
S&P 500: growth, 12 months+20.7pptrailing earnings growth · to 16 Nov 2026
S&P 500: revaluation, 12 months−10.5ppthe multiple; of which the 10-year −30.7pp (model-dependent) · to 16 Nov 2026
By accounting identity: income, growth and revaluation

In brief

The S&P 500 and the long bond taken apart by how they pay — income, growth and revaluation — and how many assets the data allow that for
  • Over the last year the S&P 500 returned +11.3%, with earnings growth adding +20.7pp.→ Exhibit 1
  • Over the last year a Treasury bond of each fund's maturity tracks the fund within 0.6pp.→ Exhibit 6
  • 12 of the 58 assets have an identity of their own; the rest are taken apart on Factor analysis.→ Exhibit 6
What would change this view
What would change this viewTriggerTodayDecided by
The long-bond model stops tracking its fund12-month outside −1.7 to +1.7pp (its own 5th–95th)−0.63pp (16 Nov 2026)not crossedthe close of 17 Nov 2026
The multiple, not earnings, carries the S&P 500's year|revaluation| > |earnings growth| over 12 months−10.5pp against +20.7pp (16 Nov 2026)not crossedthe Valuation page's next month-end update
The reasoning behind each point
  • equity
    Over the last year the S&P 500 returned +11.3%, with earnings growth adding +20.7pp.

    Revaluation, the multiple, took −10.5pp and dividends added +1.1pp. The multiple is one block; split by the valuation identity it is the 10-year's −30.7pp, as the yield moved from 4.11% to 5.12% over the page's year (252 sessions to 16 Nov 2026, +101bp) — on the register's rule, the same day a year earlier to 16 Nov 2026, +98bp, and the premium's +20.2pp. Both are model-dependent: the premium is what closes the identity, and the two are large and offsetting because the trailing spread they divide, k ÷ P/E, was only 3.0% at the start.

  • bonds
    Over the last year a Treasury bond of each fund's maturity tracks the fund within 0.6pp.

    For the 20-year-plus fund the rate move, a revaluation, took −11.2pp against +4.8pp of carry and ±0.0pp of roll-down; the 25-year bond returned −6.4% against the fund's −7.0%, a tracking gap of −0.6pp, and at 20 to 30 years it would have returned −6.7% to −5.3% (an assumption range).

  • coverage
    12 of the 58 assets have an identity of their own; the rest are taken apart on Factor analysis.

    An identity needs a price for what the fund holds: a Treasury curve, CPI, a corporate curve, an earnings series, a spot price or an exchange rate. Where the site carries none, the factor model on Factor analysis is the way to read the return, and the table says why.

Next: Factor analysis — the same return, factor by factor

Computed here from Yahoo Finance closes and the publishers' own files (U.S. Treasury, the Board's H.10, BLS, EIA); the factor model's split copied from Factor analysis.cross.decompose-read

The blocks of the return, by horizon

The return over one, three and twelve months and the year to date, split into income and carry, growth and roll, and revaluation, with the gap the identity leaves to the fund

Exhibit 1 · S&P 500 return split into income, growth and revaluation by horizon

pp of total return for one month, three months, twelve months and year to date, to 16 Nov 2026; diamond, observed return; grey, tracking gap
About
About this exhibit
Stacked bars of the S&P 500's return split into income and carry (dividends), growth and roll (earnings growth), revaluation (the multiple) and the tracking gap, with the observed return as a diamond, for each window.
S&P 500: what the return is made of, window by window
S&P 500: dividends, earnings growth and the multiple (of which the 10-year and the premium, model-dependent). The blocks are coloured by family, the same colour for the same kind of return on every asset; the grey block is the tracking gap, the fund's return less the identity's. Each window runs from the close before its first session to its last close. Over 12 months (13 Nov 2025 to 16 Nov 2026): income and carry +1.1pp, growth and roll +20.7pp, revaluation −10.5pp, tracking gap ±0.0pp; observed +11.3%.
Source: Computed here from Yahoo Finance closes and the publishers' own files (U.S. Treasury par, real and HQM curves; the Board's H.10; BLS CPI-U; EIA WTI spot; the S&P 500 earnings yield the Valuation page reads).cross.decompose-1

Exhibit 2 · S&P 500 return by source and horizon, with window dates

pp of return over one, three and twelve months and the year to date, to 16 Nov 2026; model-dependent parts marked; observed return in %
About
S&P 5001M3M12MYTD
From the close of16 Oct 202618 Aug 202613 Nov 202531 Dec 2025
To the close of16 Nov 202616 Nov 202616 Nov 202616 Nov 2026
Income and carry (dividends)±0.00pp+0.24pp+1.13pp+0.81pp
Growth and roll (earnings growth)±0.00pp−1.71pp+20.66pp+19.64pp
Revaluation (the multiple)−3.05pp−1.88pp−10.49pp−11.09pp
of which the 10-year (model-dependent)−0.92pp−11.71pp−30.70pp−28.74pp
of which the premium (model-dependent)−2.13pp+9.83pp+20.20pp+17.65pp
Tracking gap±0.00pp±0.00pp±0.00pp±0.00pp
Observed return−3.05%−3.34%+11.29%+9.36%
About this exhibit
A table of the S&P 500's return by part (income and carry, growth and roll, revaluation with its 10-year and premium sub-parts, tracking gap) and the observed return for each window, with each window's first and last close.
The blocks by horizon, with each window's dates
Percentage points of the return, on each window's simple-return scale, by block family; the identity's own parts beneath each family, the model-dependent ones marked. The observed return in percent.
Source: Computed here from Yahoo Finance closes and the publishers' own files (U.S. Treasury par, real and HQM curves; the Board's H.10; BLS CPI-U; EIA WTI spot; the S&P 500 earnings yield the Valuation page reads).cross.decompose-2

How the blocks added up month by month, and how close the identity is to the fund

The blocks month by month with the fund's own return beside them, and the identity's tracking record since the fund began

Exhibit 3 · S&P 500 monthly return split into income, growth and revaluation

Each calendar month, pp, Feb 2005 to Nov 2026; bars, income, growth, revaluation and tracking gap; line, the observed return; the last may be partial
About this exhibit
Monthly stacked bars of income and carry, growth and roll, revaluation and the tracking gap, with the S&P 500 fund's observed monthly return as a line, the most recent months shown by default.
S&P 500 month by month: the blocks and the fund's return
Each calendar month its own window, from the last close of the month before: the blocks, coloured by family, add to the observed return (line), tracking gap included. Feb 2005 to Nov 2026; the last month may be partial.
Source: Computed here from Yahoo Finance closes and the publishers' own files (U.S. Treasury par, real and HQM curves; the Board's H.10; BLS CPI-U; EIA WTI spot; the S&P 500 earnings yield the Valuation page reads).cross.decompose-3

Exhibit 4 · Monthly gap between each fund's return and its accounting split

Observed less modelled return, pp, each month the split prices the fund, to Nov 2026; none for the S&P 500, whose premium closes the gap

The premium closes the identity, so the modelled return is the observed one in every window and month: there is no tracking gap and no record to keep. What the identity can get wrong is an input — the earnings index or the payout.

About this exhibit
For an asset whose accounting split leaves a gap, the monthly difference between the fund's return and the identity's; for the S&P 500 a note that the premium closes the identity, so there is no tracking gap or record.
How close is the identity to the fund?
Every month since the identity can price the fund whole. A large gap is information — the fund holds what the identity does not — not an error to be tuned away.
Source: Computed here from Yahoo Finance closes and the publishers' own files (U.S. Treasury par, real and HQM curves; the Board's H.10; BLS CPI-U; EIA WTI spot; the S&P 500 earnings yield the Valuation page reads).cross.decompose-4

The identity beside the

The same return split by accounting identity and by the factor model over the same horizons — two readings of one return, with what each leaves

Exhibit 5 · S&P 500 tracking gap and factor-model residual, by horizon

pp: the split's tracking gap and the factor model's residual (one, three and twelve months and year to date); its share of daily variance explained, %
About
S&P 5001M3M12MYTD
Identity: window, close to close16 Oct 2026 – 16 Nov 202618 Aug 2026 – 16 Nov 202613 Nov 2025 – 16 Nov 202631 Dec 2025 – 16 Nov 2026
Identity: observed return−3.05%−3.34%+11.29%+9.36%
Identity: modelled return−3.05%−3.34%+11.29%+9.36%
Identity: tracking gap±0.00pp±0.00pp±0.00pp±0.00pp
Factor model: window, close to close16 Oct 2026 – 16 Nov 202618 Aug 2026 – 16 Nov 202613 Nov 2025 – 16 Nov 202631 Dec 2025 – 16 Nov 2026
Factor model: observed return−3.05%−3.34%+11.29%+9.36%
Factor model (the set with the market): residual−0.16pp+0.32pp+1.02pp+0.86pp
Factor model: share of daily variance explained100%100%99%99%
Factor model: below the explained-share thresholdnononono
About this exhibit
A table for the S&P 500 over each window: the identity's observed and modelled return and tracking gap beside the factor model's residual, share of daily variance explained and threshold flag.
Two readings of one return: the identity and the factor model
The identity takes the return apart by how the asset pays; the factor model by the market-wide prices it moved with, on Factor analysis. They answer different questions, so neither one's gap is the other's error — but a return both leave largely to their residuals is one neither describes.
Source: Computed here from Yahoo Finance closes and the publishers' own files (U.S. Treasury par, real and HQM curves; the Board's H.10; BLS CPI-U; EIA WTI spot; the S&P 500 earnings yield the Valuation page reads). Copied from Factor analysis's per-asset file and checked value by value against it.cross.decompose-5

Which assets have an identity of their own

Every asset in the universe, with its identity or the reason the data do not allow one; select a row to take it apart

Exhibit 6 · Every asset: its accounting split or the reason it has none

All 58 assets, twelve months to 16 Nov 2026: observed return %, the split's gap and the factor model's residual in pp, factor share explained %
AssetGroupIdentityObserved, 12MIdentity's gapFactor residualFactor share explained
IndexesS&P 500: dividends, earnings growth and the multiple (of which the 10-year and the premium, model-dependent)+11.3%±0.00pp+1.02pp99%
IndexesNo identity — The site reads no earnings history for this index, so its return cannot be split into dividends, earnings and the multiple.+18.3%–−3.05pp92%
IndexesNo identity — The site reads no earnings history for this index, so its return cannot be split into dividends, earnings and the multiple.+9.3%–+3.95pp97%
IndexesNo identity — The site reads no earnings history for this index, so its return cannot be split into dividends, earnings and the multiple.+16.4%–−8.13pp84%
IndexesNo identity — The site reads no earnings history for this index, so its return cannot be split into dividends, earnings and the multiple.+10.2%–−4.63pp89%
IndexesNo identity — The site reads no earnings history for this index, so its return cannot be split into dividends, earnings and the multiple.+14.1%–−5.18pp92%
IndexesNo identity — The site reads no earnings history for this index, so its return cannot be split into dividends, earnings and the multiple.+10.8%–+2.15pp96%
IndexesNo identity — The site reads no earnings history for this index, so its return cannot be split into dividends, earnings and the multiple.+11.1%–+5.85pp79%
IndexesNo identity — The site reads no earnings history for this index, so its return cannot be split into dividends, earnings and the multiple.+9.3%–+9.27pp72%
IndexesNo identity — The site reads no earnings history for this index, so its return cannot be split into dividends, earnings and the multiple.+12.9%–+5.23pp62%
IndexesNo identity — The site reads no earnings history for this index, so its return cannot be split into dividends, earnings and the multiple.−16.5%–−36.07pp13%
IndexesNo identity — The site reads no earnings history for this index, so its return cannot be split into dividends, earnings and the multiple.+11.1%–+8.36pp75%
About this exhibit
A table of every asset (the first rows shown, all on 'Show all'): group, its identity or the reason it has none, twelve-month observed return, the identity's gap, the factor model's residual and its explained share.
Every asset in the universe: 12 with an identity of their own, 46 read on Factor analysis
12-month figures. The identity's gap is its tracking gap, or what it cannot yet separate (for gold, its driver model's residual; for high yield, everything but the Treasury part); the factor model's residual is Factor analysis's. Select an asset to take it apart.
Source: Computed here from Yahoo Finance closes and the publishers' own files (U.S. Treasury par, real and HQM curves; the Board's H.10; BLS CPI-U; EIA WTI spot; the S&P 500 earnings yield the Valuation page reads). Copied from Factor analysis's per-asset file and checked value by value against it.cross.decompose-9

Appendix

What each identity is, what it assumes, and every source behind the page41 checks, all pass
About this page

Takes each asset's return apart by how it pays, with no model of what moved it: income and carry (dividends, a bond's carry), growth and roll (earnings growth, a bond's roll-down) and revaluation (the multiple, the rate move, a spot price), each block one colour on every asset, with the gap the identity leaves to the fund printed beside them.

1 · Treasury funds: a constant-maturity par bond, repriced each session

P(y;τ)=∑iCFi (1+y/200)−2τi,ct=200 1−Dt(T)∑k=12TDt(k/2)P(y;\tau) = \sum_{i} CF_i\,\big(1 + y/200\big)^{-2\tau_i},\qquad c_t = 200\,\frac{1 - D_t(T)}{\sum_{k=1}^{2T} D_t(k/2)}
rt,t+1=P(ct;τ−Δt)−100⏟carry+Pt(τ−Δt)−P(ct;τ−Δt)⏟roll-down+P′(ya) Δy⏟duration+12P′′(ya) Δy2⏟convexity+R3⏟higher orderr_{t,t+1} = \underbrace{P(c_t;\tau-\Delta t) - 100}_{\text{carry}} + \underbrace{P_t(\tau-\Delta t) - P(c_t;\tau-\Delta t)}_{\text{roll-down}} + \underbrace{P'(y_a)\,\Delta y}_{\text{duration}} + \underbrace{\tfrac12 P''(y_a)\,\Delta y^2}_{\text{convexity}} + \underbrace{R_3}_{\text{higher order}}
SymbolMeaningFromAssumed or estimated
Dt(τ)D_t(\tau)the day's discount factor, bootstrapped from the Treasury's par curve on a half-year grid (flat forwards)U.S. Treasury, par yield curve–
TTthe bond's maturity: TLT 25 years, IEF 8.5, SHY 2 — the middle of each fund's index band (an assumption; each band's ends give the range)the specassumed
ctc_tthe par coupon at t: the bond is bought at 100computed here–
ya, yby_a,\ y_bthe aged bond's yield on the old curve (P_t(τ−Δt)) and on the next (P_{t+1}(τ−Δt)); Δy = y_b − y_aNewton's method–
R3R_3what the second-order expansion leaves: P_{t+1}(τ−Δt) − P_t(τ−Δt) − duration − convexitycomputed here–

The five parts add to the repriced return exactly: carry is income, roll-down is growth and roll, and the three rate terms are revaluation; the tracking gap is the fund's observed return less the bond's. The TIPS fund uses the real par curve and adds the index ratio's daily accrual (reference CPI as the Treasury defines it), which is income.

2 · The S&P 500: the Valuation page's identity

Δln⁡TR=Δln⁡TR−Δln⁡P⏟dividends+Δln⁡E⏟earnings+Δln⁡(P/E)⏟the multiple\Delta\ln TR = \underbrace{\Delta\ln TR - \Delta\ln P}_{\text{dividends}} + \underbrace{\Delta\ln E}_{\text{earnings}} + \underbrace{\Delta\ln (P/E)}_{\text{the multiple}}
Δln⁡(P/E)=−ln⁡D0+ΔrfD0⏟the 10-year+Δln⁡(P/E)+ln⁡D0+ΔrfD0⏟the premium,D0=k(P/E)0=r0f+ERP0−g\Delta\ln (P/E) = \underbrace{-\ln\frac{D_0 + \Delta r^f}{D_0}}_{\text{the 10-year}} + \underbrace{\Delta\ln(P/E) + \ln\frac{D_0 + \Delta r^f}{D_0}}_{\text{the premium}},\qquad D_0 = \frac{k}{(P/E)_0} = r^f_0 + \mathrm{ERP}_0 - g
SymbolMeaningFromAssumed or estimated
EEtrailing earnings, monthly (the month's earnings yield × its average close); the current month the Valuation page's estimatethe Valuation page (copied and checked)–
kkpayout ratiothe Valuation pageassumed
rfr^fthe 10-year Treasury par yield, % (its change observed over the window)U.S. Treasury–
gggrowth, held fixed over the window (Δg = 0)assumedassumed

Dividends are income, earnings growth is growth, and the multiple is revaluation, printed as one block. The premium closes the identity, so the modelled return is the observed one and there is no tracking gap; the multiple's split into the 10-year and the premium is model-dependent, and when the starting spread D₀ is small the two are large and offsetting.

3 · Gold's driver model (read on Factor analysis): five drivers, re-fitted each month

Δln⁡Gm=a(m)+br(m) Δrm+bπ(m) Δπm+b$(m) Δln⁡$m+bS(m) Δln⁡Sm+bO(m) Δln⁡Om+em\Delta\ln G_m = a^{(m)} + b^{(m)}_{r}\,\Delta r_m + b^{(m)}_{\pi}\,\Delta \pi_m + b^{(m)}_{\$}\,\Delta\ln \$_m + b^{(m)}_{S}\,\Delta\ln S_m + b^{(m)}_{O}\,\Delta\ln O_m + e_m
explainedt=∑jb^j(m) Δxj,t,residualt=a^(m)/nm⏟alpha+restt\text{explained}_t = \sum_{j} \hat b^{(m)}_j\,\Delta x_{j,t},\qquad \text{residual}_t = \underbrace{\hat a^{(m)}/n_m}_{\text{alpha}} + \text{rest}_t
SymbolMeaningFromAssumed or estimated
Δr, Δπ\Delta r,\ \Delta\pichanges in the 10-year real yield and the 10-year breakeven, pointsU.S. Treasury–
$, S, O\$,\ S,\ Othe Board's broad dollar, US equities (SPY), oil (USO)Federal Reserve Board (H.10); Yahoo Finance–
a^(m), b^(m)\hat a^{(m)},\ \hat b^{(m)}OLS over the 60 months before month m, applied to each session's changes inside m; Newey–West errors (3 lags)OLSestimated
nmn_mNYSE sessions in month m: the constant is spread evenly over themcomputed here–

Only the five drivers' parts count as explained; the constant — alpha, gold's average return beyond its drivers over the 60 months before, carried forward — is printed with the residual. The fixed model, fitted to 2021-12 and held, measures the residual since.

4 · Windows on the simple-return scale

ℓj=∑t∈Wpartj,t 100ln⁡(1+rt/100)rt,R=∑t∈W100ln⁡(1+rt/100),printed partj=ℓj eR/100−1R/100\ell_{j} = \sum_{t \in W} \text{part}_{j,t}\,\frac{100\ln(1 + r_t/100)}{r_t},\qquad R = \sum_{t \in W} 100\ln(1 + r_t/100),\qquad \text{printed part}_j = \ell_j\,\frac{e^{R/100} - 1}{R/100}
SymbolMeaningFromAssumed or estimated
rtr_tthe period's simple return, %; its parts add to itcomputed here–
ℓj\ell_jpart j on the window's log scalecomputed here–

Each period's parts are put on its log scale, summed, and put on the window's simple scale, so they add to the compounded return, and a block family is the sum of its parts; the tracking gap is observed less modelled and is never forced to zero.

Exhibit 7 · The identities: what each is, and what it cannot separate

IdentityAssetsHow it is constructed
Treasury fund: a constant-maturity par bond repriced on the Treasury par curve every sessionTreasuries 20y+, Treasuries 7–10y, Treasuries 1–3yTreasury funds. Each session a par bond of constant maturity (TLT 25 years, IEF 8.5, SHY 2 — the middle of each fund's index maturity band) is bought at par on the Treasury's par curve, bootstrapped to discount factors on a half-year grid (flat forwards between points), and repriced one session later on the next curve. Carry is the aged bond at its unchanged yield; roll-down the aged bond on the unchanged curve; the rate move is the aged bond's own yield change Δy times its duration, plus ½ convexity × Δy², plus the rest of the Taylor expansion (higher order). Without the funds' holdings, the residual carries the maturity mismatch, the fees and the funds' own trading. The bond repriced at each end of the maturity band gives the assumption range beside the model.
TIPS fund: a constant-maturity real par bond repriced on the Treasury real par curve, plus the CPI accrualTIPSTIPS fund. The same bond on the Treasury's real par curve (extended flat below five years), 7.5 years, plus the index ratio's accrual: the reference CPI of each date is CPI-U (not seasonally adjusted) three months earlier, interpolated by day toward two months earlier, as the Treasury defines it. The accrual is applied to the repriced principal, so the parts add exactly. October 2025 CPI was never published; the site fills it with the geometric mean of the months either side, and the residual carries any difference from the Treasury's own fallback index.
High-yield fund: the Treasury part (a par bond on the Treasury curve); the spread, defaults and fees are not separatedHigh yieldHigh yield. Only the Treasury part is modelled — a 5-year par bond on the Treasury curve (an assumption, 3–7 years its range); the high-yield spread index is subscription data and is not used, so the rest — the spread's carry and change, defaults and fees — is one residual, and a large one.
Investment-grade fund: the Treasury part, spread carry and spread duration × the HQM spread's change, monthlyIG corporatesInvestment grade, monthly. The Treasury part is the 12-year par bond compounded over the month; spread carry is the previous month's spread × the month's length; the spread change is the corporate bond's modified duration (at the HQM par yield) × the spread's change. The spread is the HQM par yield at 12 years, recomputed from the Treasury's monthly spot curve, less the Treasury par yield at the same maturity averaged over the month — both monthly averages, which lag the fund's month-end prices by about half a month, and the residual carries that. The last month is Oct 2026, the HQM curve's latest.
S&P 500: dividends, earnings growth and the multiple (of which the 10-year and the premium, model-dependent)S&P 500S&P 500. The Valuation page's identity, on its earnings input and payout (0.84), over this page's windows: Δlog total return = dividends (total return less price return) + Δlog earnings + Δlog P/E, and Δlog P/E = −Δlog(rf + ERP − g) on the trailing basis with the 10-year's change observed, Δg = 0 and the premium closing it. Earnings enter monthly (the month's earnings yield × its average close), and the current month uses the 11 Sep 2026 estimate, so earnings do not move inside a month. The premium closes the identity, so there is no tracking gap to show.
Oil fund: WTI spot's change; the roll, the collateral and the fees are not separatedWTI crude (USO)Oil. The fund's return less WTI spot's log change is the roll, the collateral's interest and the fees. The spot series ends on 11 Nov 2026, the EIA's latest, so this model's windows end there; its level is not shown. The EIA's futures curve ends in April 2024, so the roll cannot be separated from the collateral.
Dollar fund: the six-currency basket's spot change; the carry and the fees are not separatedUS dollar bullish (UUP)The dollar. UUP holds futures on the ICE dollar basket. Its weights are read from ICE's own U.S. Dollar Index FAQ (read on 26 Sep 2026): the euro 57.6%, the yen 13.6%, the pound 11.9%, the Canadian dollar 9.1%, the krona 4.2%, the Swiss franc 3.6%, each equal to its exponent in the index formula the same file prints. The basket's spot change is made from the Board's H.10 noon rates with those weights (its level is not computed or shown); the rest is the futures' carry — the interest-rate gap — and the fees. The site does not read foreign rates, so the carry is not checked.
Currency fund: the spot rate's change; the interest and the fees are not separatedEuro (FXE), Yen (FXY)The euro and the yen. The fund's return less the currency's spot change against the dollar (H.10) is the interest the fund earns on its deposits, less the fees.
Gold: five drivers with 60-month point-in-time betas; the model's constant and the rest are its residualGoldGold. A regression of gold's monthly log change on the monthly changes of the 10-year real yield, the 10-year breakeven, the Board's broad dollar, US equities (SPY) and oil (USO), fitted over the 60 months before each month it is applied to, and applied to each session's changes inside that month; the month's constant is spread evenly over its NYSE sessions. That constant — alpha — is gold's average return beyond its drivers over the five years before, carried forward: no driver accounts for it, so it is printed with the residual, beside the rest, and only the five drivers' parts count as explained. It is a factor model of how gold moved with its drivers, not an identity and not a fair value, so it is read on Factor analysis; on Sources of return gold is its price change alone, because it pays no income. The fixed model, fitted to December 2021 and held, measures the residual since. The halves and the Johansen trace describe this model's own sample. Two gold models on the site answer two questions. The fair-value board's — gold change model estimated here (real yield, dollar, breakeven), fitted once on the whole sample and carried to a level, grade C — asks where gold would sit if those relations held, and its grade rests on its own halves, record and Johansen test, printed on the board. This page's re-fits five drivers on the 60 months before each month, to ask what moved gold's return over each window with the betas known at the time. They differ in drivers, sample and purpose, so neither is evidence for the other.
Source: Computed here from Yahoo Finance closes and the publishers' own files (U.S. Treasury par, real and HQM curves; the Board's H.10; BLS CPI-U; EIA WTI spot; the S&P 500 earnings yield the Valuation page reads).cross.decompose-11

Every decomposition here is a description of sessions that have happened, not a forecast. Each native model is a model of the fund's return made from a price of what the fund holds; its parts add to the modelled return exactly, and the tracking gap — the observed return less the modelled one — is printed beside them and never forced to zero. Windows are Factor analysis's: the last 21, 63 and 252 sessions and the year to date, ending on the same day, so the observed return is the same number on both pages. Parts are put on each window's simple-return scale ((e^R − 1) ÷ R), so they add to the compounded return.

The factor model's split, printed in Detail beside each identity, is copied from Factor analysis's per-asset file, on the asset's default factor set, and checked value by value against that file: the 3-month bill, each factor's contribution (a point-in-time beta × the factor's move) and the residual. Nothing in it is recomputed here.

The page carries the S&P 500 whole; every other asset's windows, months and factor split are in a file of its own, loaded when it is selected.

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.