Δlog oil vs Δyield (FT replica)
\[\begin{aligned} x_t &= \ln P_t - \ln P_{t-1} \\ z_t &= y_t - y_{t-1} \end{aligned}\]The oil return (log %) against the daily change in the yield (pp). It captures day-to-day co-movement.
Government borrowing costs are at the mercy of oil prices
Oil/10-year Treasury, three-month rolling correlation (%)
Key points
- Replication. The FT chart is the correlation of daily WTI log returns with daily changes in the 10Y yield, over a rolling 63-trading-day (3-month) window. It matches the FT's landmarks: +70 in late 1990 (Gulf War), −54 in early 2003 (Iraq war), +63 in 2026.
- Gaps vs FT. FT starts in 1985 and looks slightly noisier, so it probably uses futures. FRED's WTI spot series starts in 1986.
- Hypothesis (Erdem Hoca). The oil–yield link depends on the regime. Above some threshold the correlation switches on. Next step: a Hansen (2000) threshold model with the threshold γ estimated and bootstrap inference.
- Warning. High oil–yield correlation shows up in two different regimes. In 2009–14 it was demand-driven: oil, stocks and yields all moved together. From 2022 it is supply/inflation-driven: oil and yields rise while stocks fall. A threshold on the oil price alone may lump the two together. Inflation or breakevens are strong alternative threshold variables (see the Stock–Bond tab).
- "Gap vs avg" transform. Oil: ln(WTI) minus its trailing 12-month average. Bond: the 10Y yield level minus its trailing 12-month average, in pp; no log is needed for yields. It captures how far each market is from its recent norm, which is cycle-level co-movement rather than day-to-day reaction. Caveat: the gaps are very persistent (AR(1) ≈ 0.99), so a 3-month correlation swings between −90 and +98 (SD 47 vs 20 for returns), with |corr| > 50 in 44% of windows vs 4%. Use 12-month or longer windows with it. Its full-sample correlation is 0.38, and its 3-month version is higher than the FT measure in both 2009–14 (39 vs 25) and 2022–26 (36 vs 22).
- LOESS (span 0.3, FT transform). The full-sample curve is an inverted U: the correlation rises from about 5% at $20 to about 23% around $80, then falls to about 5% near $115. That is not a simple "higher oil → stronger link". By period, the 2022+ curve falls from about 40% at $60 to about −5% above $110, while 2000–21 is flat at 10–25%. Whatever switches the regime does not look like the oil price level alone, which is consistent with inflation being the stronger threshold variable in the Research tab. The extremes rest on few, highly overlapping windows.
- Descriptive split (3m window). Mean oil–bond correlation is 0.07 pre-2000, 0.15 in 2000–21 and 0.22 from 2022.
Oil
Window (months)
63 trading days
Transform
Gap: average length
Frequency
Period
–
Methods & notation: what each setting computes
\(P_t\): oil price (WTI or Brent spot, $/bbl) on trading day \(t\); \(y_t\): 10-year Treasury yield (%). Only days with both series observed and \(P_t > 0\) are used. Each transform turns \((P_t, y_t)\) into a pair \((x_t, z_t)\), and the chart plots the rolling correlation of \(x\) and \(z\).
% change
\[\begin{aligned} x_t &= P_t / P_{t-1} - 1 \\ z_t &= y_t / y_{t-1} - 1 \end{aligned}\]Simple percentage changes of both. For the yield this is a relative change, which inflates moves when yields are low.
Levels
\[\begin{aligned} x_t &= P_t \\ z_t &= y_t \end{aligned}\]Raw levels. Both are near-unit-root series, so the correlation is prone to being spurious.
Gap vs avg
\[\begin{aligned} x_t &= \ln P_t - \tfrac{1}{M}\textstyle\sum_{j=0}^{M-1} \ln P_{t-j} \\ z_t &= y_t - \tfrac{1}{M}\textstyle\sum_{j=0}^{M-1} y_{t-j} \end{aligned}\]The deviation from a trailing average that includes day \(t\). \(M = 21 \times\) (average length in months), e.g. \(M = 252\) for 12 months. Oil in logs, yield in level (pp). The series are very persistent (AR(1) ≈ 0.99).
Rolling correlation (window \(W\))
\[\rho_t = 100 \times \frac{\sum_{s \in \mathcal W_t} (x_s - \bar x_t)(z_s - \bar z_t)}{\sqrt{\sum_{s \in \mathcal W_t} (x_s - \bar x_t)^2 \,\sum_{s \in \mathcal W_t} (z_s - \bar z_t)^2}}\] \[\mathcal W_t = \{t-W+1, \dots, t\}, \qquad \bar x_t = \tfrac{1}{W}\textstyle\sum_{s \in \mathcal W_t} x_s\]A Pearson correlation over the last \(W\) observations, dated at the window's last day (trailing, no look-ahead). Daily: \(W = \operatorname{round}(21 m)\), so 3 months = 63 days. Weekly: \(W = \operatorname{round}(\tfrac{52}{12} m)\), so 3 months = 13 weeks. With "Gap vs avg" the first value needs \(M + W - 1\) days.
Weekly frequency
\(P_w, y_w\) are the last available observations in the week ending Friday. The transform is then applied to the weekly series, e.g. \(x_w = \ln P_w - \ln P_{w-1}\).
Oil threshold split
Each window is assigned by the oil price on its last day: high if \(P_t \ge c\), low otherwise. The tiles show \(\frac{1}{|\{t: P_t \ge c\}|}\sum_{t: P_t \ge c} \rho_t\) and its complement. This is descriptive only; the formal test is in the Research tab.
Bin means
\[\bar\rho_k = \operatorname{mean}\{\rho_t : 10k \le P_t < 10(k+1)\}\]Bins of $10, shown only if they contain at least 20 windows.
LOESS
\[\begin{aligned} \hat g(p_0) &= \hat a, \quad (\hat a, \hat b) = \arg\min_{a,b} \textstyle\sum_i w_i e_i^2 \\ e_i &= \rho_i - a - b\,(P_i - p_0) \end{aligned}\]Tricube weights \(w_i = (1 - |P_i - p_0|^3 / h^3)^3\) for \(|P_i - p_0| < h\), where \(h\) is the distance to the \(\lceil \text{span} \cdot n \rceil\)-th nearest \(P_i\). Evaluated between the 1st and 99th percentile of \(P\).
Is the correlation higher when oil is expensive?
Each point is one day: the rolling correlation against that day's oil price. The line is the mean correlation within $10 oil-price bins. This is a descriptive preview, not a formal threshold test.
LOESS: a smooth alternative to the bins
The same points with a LOESS curve: a local linear fit with tricube weights (Cleveland 1979). The span is the share of observations used in each local fit; smaller spans follow the data more closely. "By period" fits a separate curve for each regime, to show whether the oil-price/correlation relationship itself shifts. The curve is descriptive: overlapping windows make nearby points strongly dependent, so no confidence band is shown.
Span
Curves
Stock–bond correlations
Correlation of daily stock returns with daily changes in 10-year government bond yields
Key points
- Paper. Canlin Li & Emre Yoldas, "Government Bonds' Safe-Haven Property and Inflationary Pressures" (link). Whether bonds hedge stocks depends on the inflation environment and the type of shock. With low, anchored inflation, central banks ease after both demand and supply shocks, so bonds hedge. With inflation pressure (post-Covid), adverse supply shocks bring expected tightening, and the hedge breaks. The post-2021 deterioration reflects the environment, not a permanent change in bonds.
- Sign convention. Stock returns vs. yield changes. Positive means bonds hedge stocks (yields fall when stocks fall). Negative means stocks and bond prices fall together.
- Link to our oil–bond work. An oil spike is a classic adverse supply shock. In an inflationary regime, the paper's mechanism predicts oil–yield correlation ↑ and stock–yield correlation ↓. So the inflation regime is a natural threshold variable for the Hansen model, alongside the oil price.
- What the data says (US, 12m window). Stock–bond means are −0.34 pre-2000, +0.32 in 2000–21 and −0.06 from 2022. Oil–bond means are 0.09, 0.15 and 0.19. Oil–bond is highest from 2022 ✔︎. But pre-2000 had low oil–bond correlation despite negative stock–bond, so the story is not just inflation. Within 2000–21 the two correlations co-move (ρ≈0.5): the demand channel.
- Replication quality. The US and UK match Figure 2 closely; the US latest is about −0.35 and the UK about −0.6. Euro area uses free proxies: DAX or Euro Stoxx 50 for stocks, and the ECB AAA 10Y (from 2004) for the yield. That matches the end-2026 drop to about −0.55. The Bundesbank Bund series (from 1997) is recorded earlier in the day than the stock close, which biases the correlation down, and its updates lag.
Window (months)
252 trading days
Euro-area equity
Euro-area 10Y yield
UK 10Y yield
Regime shading
Period
–
Methods & notation
\[\rho^{SB}_t = \operatorname{corr}_{s \in (t-W, t]}\big(\ln S_s - \ln S_{s-1},\; y_s - y_{s-1}\big), \qquad W = \operatorname{round}(21 m) \;(252 \text{ for 12 months})\]\(S\): equity index (S&P 500, DAX/Euro Stoxx 50, FTSE 100); \(y\): that country's 10Y yield. Each country uses its own trading days where both are observed. A positive value means bonds hedge equities. The regime table reports the mean of \(\rho^{SB}_t\) by calendar period. The oil–bond row uses \(x_t = \ln P_t - \ln P_{t-1}\) with the same \(W\).
Oil–bond vs. stock–bond, same window
If oil spikes act as inflationary supply shocks, the oil–bond line should rise when the stock–bond line falls (as from 2022). When the two move together, demand and growth news is driving all three markets (as in 2009–14).
Working paper design · draft
When do oil shocks move Treasury yields? Inflation regimes and the oil–bond link
A threshold-regression study of state-dependent oil-price transmission to long-term government bond yields
Key points (preliminary, US 1986–2026)
- The oil–yield slope is state-dependent. When only the oil slope is allowed to switch, Hansen's sup-Wald test rejects linearity for the oil price (nominal and real) and for CPI inflation, at both daily and weekly frequency. The slope is about 0.1–0.2 bp per 1% oil move in the low regime and about 0.4–0.6 bp in the high regime: 2–4× stronger.
- Inflation threshold ≈ 3% y/y. γ̂ = 3.1% at both daily and weekly frequency. The main 95% confidence region is about [2.7, 3.7]% (daily) and [2.7, 4.0]% (weekly); the daily LR profile also has a small isolated region near 1.4–1.5%. It is the only variable that is significant and stable at both frequencies. Current CPI inflation (3.65%) is above the threshold, consistent with the 2026 spike in the FT chart.
- Not a "price of oil" story only. The horse race on a common sample gives no clear winner. Nominal oil fits best on daily data, real oil on weekly data, and the R² gains are small (<1 pp). The nominal oil price trends with time, so its threshold may partly be a structural break. That is the main identification issue to address.
- Breakeven inflation (2003+) and oil relative to its trend are not significant as threshold variables in the oil-slope-only model.
- The full sample split is dominated by the S&P 500 coefficient, i.e. the stock–bond regime. That is why the headline test must isolate the oil slope.
Research question
Under what state of the economy does an oil-price shock move long-term government yields, and which state variable governs the switch: the oil price itself, realized inflation, or inflation expectations?
Contribution
- Estimated regimes, not imposed ones. Prior work on bond–equity and bond–inflation comovement fixes regimes by date or lets them vary smoothly. We estimate a threshold and infer it formally (Hansen 1996, 2000).
- A horse race of state variables: oil price vs. inflation vs. expectations, nested in one framework.
- Mechanism. We separate supply-driven from demand-driven oil moves (Kilian 2009; Ready 2018; Känzig 2021) to test the policy-reaction channel proposed by Li & Yoldas.
- External validity across the US, UK and euro area.
Hypotheses
H1 · Threshold
The response of the 10Y yield to oil returns differs across regimes defined by a predetermined state variable qt−1; linearity is rejected.
H2 · Inflation state
The relevant state is the inflation environment. An inflation-based threshold fits at least as well as an oil-price threshold and survives controls for time variation.
H3 · Policy channel
In the high regime the stronger response comes from supply-driven oil shocks and runs through expected policy rates (2Y yields, breakevens). In the same regime the stock–bond correlation turns negative.
Empirical model
Δyt = α + β1 roilt · 1{qt−1 ≤ γ} + β2 roilt · 1{qt−1 > γ} + δ reqt + εt
- Variables. Δyt is the change in the 10Y Treasury yield (bp). roil is the WTI log return (%) and req is the S&P 500 log return (%), which controls for common demand and risk news.
- Threshold variables, lagged one period: nominal and real WTI; CPI y/y, available from mid-month t+1 so there is no look-ahead; 10Y breakeven; WTI relative to its 1-year average.
- Estimation. γ̂ minimizes SSR over all admissible splits, with 15% trimming. H0: β1 = β2 is tested with a heteroskedasticity-robust sup-Wald statistic and Hansen's (1996) fixed-regressor wild bootstrap (499 draws). The CI for γ inverts LR(γ)/η̂² at 7.35 (Hansen 2000).
- Robustness spec. The same model with all coefficients switching (a full sample split).
Preliminary results
Specification
Frequency
Plot threshold variable
*, **, *** significant at 10%, 5%, 1% (bootstrap). βoil is in bp of 10Y yield per 1% oil return. Click a row to plot it.
LR profile
Threshold variable over time
Horse race: which state variable splits the sample best?
Identification & robustness plan
Threats
- Time trend as threshold. Nominal oil and inflation are persistent, so a threshold may be a disguised structural break. Re-estimate within sub-samples, add Bai–Perron break dummies, and use detrended or real q.
- Oil endogeneity. Oil and yields both react to global demand news. Use identified oil supply news (Känzig 2021) or Ready's (2018) supply/demand decomposition, and an IV threshold model (Caner & Hansen 2004).
- Daily dependence. Add block or wild-cluster bootstrap and HAC standard errors; check weekly and monthly frequencies.
Extensions
- Decomposition: 10Y nominal = real (TIPS,
DFII10) + breakeven; 2Y vs 10Y (policy expectations vs term premium). - Joint model with the stock–bond correlation: does the high oil regime coincide with a negative stock–bond correlation?
- International panel: UK and euro-area yields with Brent.
- Out-of-sample: does the regime predict the next quarter's oil–yield correlation?
Positioning & literature
| Reference | Role in the paper |
|---|---|
| Hansen, B. E. (2000). Sample splitting and threshold estimation. Econometrica 68(3), 575–603. doi | Estimator, LR confidence set for γ |
| Hansen, B. E. (1996). Inference when a nuisance parameter is not identified under the null hypothesis. Econometrica 64(2), 413–430. doi | sup-Wald test, bootstrap p-values |
| Caner, M. & Hansen, B. E. (2004). Instrumental variable estimation of a threshold model. Econometric Theory 20(5), 813–843. doi | Endogenous oil returns |
| Kilian, L. (2009). Not all oil price shocks are alike. American Economic Review 99(3), 1053–1069. doi | Supply vs. demand oil shocks |
| Kilian, L. & Park, C. (2009). The impact of oil price shocks on the U.S. stock market. International Economic Review 50(4), 1267–1287. doi | Asset-price response by shock type |
| Ready, R. C. (2018). Oil prices and the stock market. Review of Finance 22(1), 155–176. doi | Daily supply/demand decomposition |
| Känzig, D. R. (2021). The macroeconomic effects of oil supply news. American Economic Review 111(4), 1092–1125. doi | Identified oil supply news |
| Bernanke, B. S., Gertler, M. & Watson, M. (1997). Systematic monetary policy and the effects of oil price shocks. Brookings Papers on Economic Activity 1997(1), 91–157. doi | Policy-reaction channel |
| Hamilton, J. D. (2003). What is an oil shock? Journal of Econometrics 113(2), 363–398. doi | Nonlinear oil-price effects |
| Kilian, L. & Vega, C. (2011). Do energy prices respond to U.S. macroeconomic news? Review of Economics and Statistics 93(2), 660–671. doi | Daily timing assumptions |
| Baele, L., Bekaert, G. & Inghelbrecht, K. (2010). The determinants of stock and bond return comovements. Review of Financial Studies 23(6), 2374–2428. doi | Stock–bond comovement drivers |
| Campbell, J. Y., Sunderam, A. & Viceira, L. M. (2017). Inflation bets or deflation hedges? Critical Finance Review 6(2), 263–301. doi | Time-varying bond risk, inflation regimes |
| Campbell, J. Y., Pflueger, C. & Viceira, L. M. (2020). Macroeconomic drivers of bond and equity risks. Journal of Political Economy 128(8), 3148–3185. doi | Monetary-policy regimes and the bond–stock correlation |
| Li, C. & Yoldas, E. Government bonds' safe-haven property and inflationary pressures. link (venue and year to be confirmed) | Motivation: inflation regime and shock type |
All references except Li & Yoldas were checked against Crossref (DOI, volume, pages).
Target journals
Reach
Journal of Monetary Economics, Review of Economics and Statistics, Review of Financial Studies. These need the mechanism (H3) nailed down with identified supply shocks and the policy-expectations decomposition, plus international evidence.
Strong fit
Journal of International Economics, Journal of Financial and Quantitative Analysis, Journal of Money, Credit and Banking. These need a clean estimated threshold, robust identification and a convincing horse race.
Field
Energy Economics, Journal of International Money and Finance, Journal of Banking & Finance. The current preliminary results plus the robustness plan would suffice.
Research 2 · paper proposal and evidence · draft
Anchored or not: oil supply shocks, inflation regimes and the Treasury yield curve
Oil supply shocks move Treasury yields only when inflation is away from target, and the channel switches between reflation and stagflation
Key points
- Headline. Daily oil returns are split into supply, demand and risk shocks (Ready 2018). When CPI inflation is within about ±1.1 pp of 2%, oil supply shocks have no effect on the 10Y yield (β ≈ 0.04 bp per 1%). Outside that band the effect is about 3–9× larger (β ≈ 0.37–0.38 vs 0.04–0.11 depending on outlier treatment; sup-Wald bootstrap p ≤ 0.002, Table 3). Demand shocks move yields in both regimes.
- The relationship is U-shaped, not "high inflation only". The pre-specified state \(|\pi_{t-1}-2|\) beats plain inflation. A data-driven two-sided band, with no centre imposed, lands at 1.4%–3.1%, i.e. around the target.
- Not a Covid or time artefact. The target-gap threshold is significant in every subsample: pre-Covid 1990–2019, 1990–2007, 2000– and 2008– (p ≤ 0.002, βsup ≈ 0 near target in all of them). A plain inflation threshold is not stable (pre-Covid p = 0.13).
- It is not the oil price. With oil-price states, the supply-slope test is insignificant (p = 0.15 real, 0.37 nominal); the earlier oil-level "threshold" came from demand shocks and time.
- Mechanism (TIPS, 2003–). Above the band, supply shocks raise breakevens (+0.45) and lower real yields (−0.28): stagflation pricing. Below the band, real yields and breakevens move with oil (+0.23, +0.25): reflation/deflation pricing. Near target, nothing moves.
- External identification (Känzig 2021), mixed. On OPEC announcement days the IV estimates match the story almost exactly: the 10Y moves 0.02 bp per 1% oil move near target vs 0.85 far from target, breakevens 0.33 vs 0.96. But with 143 events the differences are only marginal (t = 1.5 and 1.9). The near-target 10Y response stays at or below zero for 20 trading days. In monthly 2SLS with the VAR-based news shock there is no regime difference (0.58 vs 0.77). Current reading: anchoring shows up in how markets reprice on impact; whether it survives at monthly horizons is open.
- Link to tab 2. Using the lagged stock–bond correlation as the state gives the same picture from the other side: supply shocks matter when the stock–bond correlation is low or negative (β 0.26 vs 0.01), the Li & Yoldas inflation-regime signature.
What tabs 1–3 established
1 · Oil–Treasury (FT)
The FT chart is replicated (Δlog WTI vs Δ10Y, 63 days). The correlation rose after 2022, but LOESS shows an inverted U in the oil price, and after 2022 the link even weakens at very high prices. The oil price level is not the switch.
2 · Stock–Bond
The Li & Yoldas Figure 2 is replicated for the US, UK and euro area. The bond hedge flips with the inflation regime. High oil–yield correlation appears in both demand-driven (2009–14, stock–bond > 0) and supply/inflation-driven (2022+, stock–bond < 0) episodes.
3 · Research
The Hansen threshold model shows the oil slope roughly doubles above about 3% CPI. But oil-price thresholds may be time effects, and the horse race had no clear winner. Without separating supply and demand shocks there is no mechanism.
→ Research 2
Separate the shocks, let the distance from target govern the regime, and trace the channel through 2Y, 10Y, real and breakeven yields. That turns a correlation pattern into an economic mechanism.
Paper pitch
Question & answer
Q. When do oil supply shocks move long-term government yields, and why?
A. Only when inflation is away from target. There the shock changes expected inflation and policy, with opposite real-rate signs above and below the band. Near target, expectations are anchored and bond markets look through supply shocks.
Contribution
- An estimated anchoring band for the pass-through of oil shocks to bond yields, with Hansen-type inference and an endogenous band. To our knowledge this is new; a systematic literature check is still pending.
- It reconciles the time-varying oil–bond correlation (FT) and stock–bond correlation (Li & Yoldas; Campbell, Pflueger & Viceira 2020) in a single state variable.
- It decomposes the channel across the curve: real vs breakeven, short vs long. Stagflation and reflation pricing emerge as two sides of de-anchoring.
Methods & notation
1 · Shock decomposition (Ready 2018)
\[\begin{aligned} u_t &= \text{VIX}_t - \hat E[\text{VIX}_t \mid \text{VIX}_{t-1}, \text{VIX}_{t-2}] \\ r^{P}_t &= a_1 + b_1 u_t + d_t \\ r^{O}_t &= a_2 + c_d\, d_t + c_u\, u_t + s_t \end{aligned}\]\(r^P\): NYSE Arca Oil index return (oil producers); \(r^O\): WTI log return (%). Components: supply \(s_t\), demand \(c_d d_t\), risk \(c_u u_t\). They sum to the demeaned oil return and are mutually orthogonal by construction.
2 · Regime model
\[\Delta y_t = \alpha + \sum_{k \in \{s,d\}} \big(\beta_k^{N} \mathbb 1_{q_{t-1} \le \gamma} + \beta_k^{F} \mathbb 1_{q_{t-1} > \gamma}\big)\, o^k_t + \delta\, o^{u}_t + \varepsilon_t\]Main state \(q_{t-1} = |\pi_{t-1} - 2|\), with CPI y/y available from mid-month \(t+1\). γ is estimated by least squares; the sup-Wald test of \(\beta_s^N = \beta_s^F\) uses Hansen's (1996) fixed-regressor wild bootstrap; the confidence set inverts LR(γ). Δy is in bp; \(o^k\) in % oil-return units.
3 · Channels & robustness
Same regressors for \(\Delta y \in\) {2Y, 10Y, 10Y−2Y, 10Y TIPS, 10Y breakeven}, with three regimes: below \((\pi \le 2-\hat\gamma)\), near, above \((\pi > 2+\hat\gamma)\). Robust (HC0) SEs. The band is also estimated without a centre: \(\min_{g_1 < g_2} \text{SSR}\) with "inside" \(= \{g_1 < \pi \le g_2\}\). Rolling 252-day OLS betas give the time profile.
Evidence
Table 1 · Which state variable governs supply-shock pass-through?
β in bp of 10Y yield per 1% oil move driven by that shock; robust SEs in parentheses. "low" = q ≤ γ̂, "high" = q > γ̂ (for the stock–bond state, high = positive correlation). p (supply) tests only βsup. *R² gain vs. the linear model on the common sample from the race (stock–bond requires a year of history). Click a row to plot its LR profile.
LR profile
Table 2 · Stability across subsamples
Table 3 · Outlier robustness (WTI went negative on 2020-04-20; daily |Δln P| reached 72%)
Figure 1 · Through which part of the curve? Responses by inflation regime
Shock
Sample
"Below target band" comes from a few deflation-scare episodes (2009, 2015, 2020); read it with care.
Figure 2 · Time-varying pass-through (rolling 1-year betas)
Shaded: weeks when CPI inflation is outside the estimated target band. Betas come from winsorized oil returns (0.5%/99.5%) so the April 2020 negative-price days do not dominate a whole year of windows.
Step 2 · External identification with OPEC supply news (Känzig 2021)
The Ready decomposition is a statistical split. Here the supply shock is instead identified from futures-price surprises on OPEC announcement days (daily) and from Känzig's VAR-based oil supply news shock (monthly), and the oil return is instrumented by them separately in each regime.
Design
IV β in bp of yield per 1% oil move caused by the identified supply shock; robust SEs in parentheses. Regimes use γ̂ from the main model (|π − 2| ≤ 1.12 pp = near target).
Figure 3 · How the response unfolds after an OPEC surprise
Variable
Cumulative change from the close before the announcement to day h, per unit of surprise; bands ±1.96 robust SE.
Draft paper outline
- Introduction. The FT fact; the stock–bond analogue; our answer (an anchoring band); three findings.
- Related literature. Oil shocks and asset prices (Kilian 2009; Kilian & Park 2009; Ready 2018; Känzig 2021); monetary reaction (Bernanke, Gertler & Watson 1997); bond risk regimes (Baele et al. 2010; Campbell et al. 2017, 2020; Li & Yoldas); threshold econometrics (Hansen 1996, 2000).
- Data and shock identification. Daily 1990–2026; decomposition; summary statistics.
- Econometric framework. Threshold model, bootstrap, LR sets, band estimator.
- Main results. Table 1 (state horse race), the LR profile, Table 2 (stability).
- Mechanism. Figure 1 (curve decomposition: stagflation vs reflation pricing), Figure 2 (time profile).
- Robustness. Futures and Brent; centre 2.5%; PCE and real-time CPI vintages; weekly data; Känzig supply news as instrument; international (UK, euro area against their targets).
- Conclusion. Policy credibility as a hedge against oil-driven bond risk.
What is still needed for a top journal
Identification
- Känzig (2021) OPEC supply news: done (Step 2): direction confirmed, power limited, monthly results differ. Next: pool event days across countries; an IV threshold model (Caner & Hansen 2004).
- Front-month futures and Ready's original world producer index instead of spot WTI and XOI.
- Inference on the two-sided band itself, not only on a pre-specified centre.
Expectations data
- Real-time CPI vintages (ALFRED) and PCE as the target variable.
- Survey-based de-anchoring measures (e.g. household and professional forecasts) as an alternative state.
- Term-premium vs expected-rate split of the 10Y response.
External validity
- UK (gilts, BoE target) and euro area (ECB AAA, HICP) with Brent: does each market's own target band govern pass-through?
- Pre-1990 sample using a futures-based decomposition.
- Out-of-sample: does being outside the band predict the next quarter's oil–bond correlation?
Target journals (my judgment)
A single, clean finding with a mechanism fits Journal of Monetary Economics (first choice) or AER: Insights (a short format suits a sharp result). Then Review of Economics and Statistics or Journal of Financial and Quantitative Analysis. Reaching Review of Financial Studies or Journal of Finance would require the IV identification plus the international evidence.