## wp1914

## Source details

**Canonical URL:** [wp1914](https://www.imf.org/-/media/files/publications/wp/2019/wp1914.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wp1914.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wp1914.pdf.json)

---

### I. Introduction — context and motivation
- Definition:
  - CIP deviation x_t,t+n = (f_t,t+n − s_t) − (r*_t,t+n − r_t,t+n).
  - “Negative dollar basis” when x_t,t+n < 0.
- Historical pattern:
  - Pre-GFC: CIP held closely at daily or weekly frequencies.
  - During GFC and post-GFC: persistent CIP deviations emerged and persisted.
- Policy relevance:
  - CIP deviations may signal financial-market distortions and inefficiencies in resource allocation.
  - Failure of CIP undermines the claim that small open economies can fully insulate domestic monetary policy from U.S. Federal Reserve choices via automatic exchange-rate adjustments.
  - Understanding drivers of CIP departures is key to assessing cross-border monetary policy transmission.

### II. Literature and proposed drivers
- Proposed drivers surveyed:
  - Regulation-induced or other arbitrage limits.
  - Changes in banks’ balance-sheet capacity tied to U.S. dollar appreciation.
  - Interest-rate differences across currencies impacting the swap market.
- Empirical approach:
  - Macro-financial level analysis across ten currencies relative to the U.S. dollar.
  - Methods: time-series and panel estimation, rolling estimation windows, Markov regime-switching models, split-sample analysis.
  - Alternative funding-rate measures and an instrumental-variable design used given concerns about Libor as a marginal funding-cost proxy.

### III. Measurement and empirical observations (pre-, during-, and post-GFC)
- Measurement:
  - Horizons: 3-month and 5-year CIP deviations.
  - Alternative short-term funding-rate: 3-month U.S. AA-rated financial commercial paper (CP) rate and foreign 3-month government bill rate.
  - FX forward liquidity: forward point bid-ask spreads.
  - Capital-charge measure: VaR-based capital charge in a 5-year Libor CIP trade (Du, Tepper, and Verdelhan (2017)).
- Key empirical facts:
  - Pre-GFC: deviations very small, around zero.
  - During GFC:
    - Short-term CIP deviations reached about -200 basis points.
    - Five-year deviations were more negative than -50 basis points.
  - Post-GFC:
    - CIP deviations decreased toward zero through 2013, widened after mid-2014.
    - Volatile but persistent CIP deviations appear during and after the GFC.
    - Libor-based and CP–government-bill based bases can differ notably during the GFC; behavior more similar after the GFC.
  - Currency patterns:
    - Most currencies show negative dollar basis (x_t,t+n < 0).
    - “Carry” currencies such as AUD and NZD display positive deviations (x_t,t+n > 0) relative to USD.
  - FX liquidity and capital charges:
    - Bid-ask spreads small but show spikes with increased frequency during and after the GFC.
    - Tighter capital requirements and higher capital charges since the GFC appear to have increased arbitrage costs, but do not fully explain temporary restoration of CIP in 2013–2014 or widening after mid-2014.

### IV. Main findings and interpretation
- Principal conclusions:
  - CIP broke down during the GFC and has not held reliably since; deviations increased significantly post-GFC across benchmarks and currency pairs.
  - Structural factors (post-crisis financial regulations) likely increased arbitrage costs, but multiple drivers—some temporary—also contributed. Temporary drivers cited include asynchronous monetary policy and the October 2016 U.S. prime MMF reform.
  - Time-series evidence: factors with strong statistical associations across the sample (notably U.S. dollar strength) do not have uniform importance across currency pairs and time.
- Robust associations across maturities:
  - U.S. dollar strength, global risk sentiment (VIX), and, to a limited extent, forward exchange-market liquidity significantly contributed to cross-currency basis variation across maturities.

### V. 2.2 Basic regression analysis (Libor-based cross-currency calculations) — methodology
- Empirical specification:
  - First-difference regressions of Δ(f−s) on Δ(r*−r) and controls: Δ USDINDEX (FRB U.S. trade-weighted broad dollar index), Δ lnVIX (log change in U.S. VIX), and Δ Forward Bid-Ask (forward bid-ask spread).
  - Weekly averages for G8 currencies’ 3-month forward premiums, Libor/interbank rate differential, USDINDEX, lnVIX from 2002 to 2018 with sample splits (pre-crisis, crisis, post-crisis).
- Alternative approaches:
  - Replace Libor differential with 3-month foreign government bill − U.S. commercial paper differential.
  - IV: use commercial paper − treasury bill differential as instrument for Libor differential.

### VI. 2.2 Baseline regression results — selected numeric findings
- Pre-crisis (2002-2006, weekly, 3-month Libor deviations):
  - Canada (CAD): Δ(r*-r): 0.858*** (0.0347); N: 259; R-sq: 0.779.
  - Euro Area (EUR): Δ(r*-r): 0.978*** (0.0288); Δ Forward Bid-Ask: -0.629** (0.316); N: 259; R-sq: 0.857.
  - Panel (CAD CHF DKK EUR GBP JPY NOK SEK): Δ(r*-r): 0.976*** (0.0243); N: 2072; R-sq: 0.798.
- Crisis (2007-2009, weekly, 3-month Libor deviations):
  - Canada (CAD): Δ(r*-r): 0.920*** (0.108); N: 156; R-sq: 0.617.
  - Euro Area (EUR): Δ(r*-r): 0.820*** (0.150); Δ Forward Bid-Ask: -10.38*** (3.821); N: 156; R-sq: 0.563.
  - Panel: Δ(r*-r): 0.891*** (0.0274); Δ USDINDEX: -1.327*** (0.308); N: 1248; R-sq: 0.481.
- Post-crisis (2010-2018M6, weekly, 3-month Libor deviations):
  - Switzerland (CHF): Δ(r*-r): 1.572*** (0.207); Δ USDINDEX: -0.967* (0.573); Δ lnVIX: -6.424** (3.127); Δ Forward Bid-Ask: -3.342*** (0.709); N: 442; R-sq: 0.448.
  - Japan (JPY): Δ(r*-r): 0.898*** (0.240); Δ USDINDEX: -1.735*** (0.427); Δ lnVIX: -5.846** (2.415); N: 442; R-sq: 0.198.
  - Panel: Δ(r*-r): 1.082*** (0.186); Δ USDINDEX: -1.432*** (0.248); Δ lnVIX: -3.556** (1.357); Δ Forward Bid-Ask: -0.0557* (0.0264); N: 3536; R-sq: 0.331.

### VII. 2.2 IV regression results — selected numeric findings
- Motivation: address measurement error in Libor-based proxies.
- Pre-crisis IV (Panel CAD EUR JPY NOK SEK):
  - Δ(r*-r): 0.980*** (0.0211); Δ USDINDEX: -0.0568 (0.0930); N: 1295; R-sq: 0.814.
- Crisis IV (Panel):
  - Δ(r*-r): 1.517*** (0.133); Δ USDINDEX: 1.759*** (0.583); Δ Forward Bid-Ask: -0.367* (0.189); N: 755; R-sq: 0.651.
- Post-crisis IV (Panel):
  - Δ(r*-r): 1.101*** (0.265); Δ USDINDEX: -1.667*** (0.314); Δ lnVIX: -2.866*** (0.902); Δ Forward Bid-Ask: -0.0282** (0.0121); N: 2192; R-sq: 0.408.
- Interpretation:
  - IV results broadly support OLS qualitative findings: USD strength, VIX, and liquidity matter more post-crisis; measurement divergences across funding-rate proxies notable in stress periods (late 2008).

### VIII. 3.1 Additional hypotheses and data constructs
- Monetary policy divergence hypotheses:
  - Hypothesis 1 (IOER differential): larger IOER* − IOER leads to decreased FX hedge demand (negative coefficient expected).
  - Hypothesis 2 (Corporate issuance — Residualized_Spread): more favorable domestic borrowing conditions (lower Residualized_Spread) increase hedge demand (positive coefficient expected).
  - Data/controls include IOER differential, Residualized_Spread, Δ USDINDEX, Δ lnVIX, Δ Forward Bid-Ask.
- Prime MMF reform / dollar funding channel:
  - Hypothesis 3 (Dollar funding strain): negative shocks to prime MMF holdings (itHoldings) increase CIP deviations in favor of USD (positive coefficient expected).
  - itHoldings constructed monthly using OFR prime MMF gross holdings mapped to bank branches by headquarters; included in regressions with dollar strength and VIX.
- Domestic risk sentiment:
  - Hypothesis 4 (Domestic risk prospect): Δ lnVDAX widens EUR/USD basis; Δ lnVFTSE widens GBP/USD basis (expect negative coefficients when measured as increases widen basis in favor of USD).
  - VDAX-New and VFTSE used; residualized versions (on U.S. VIX) also employed.

### IX. 3.2 Empirical tests: selected results by hypothesis (numeric highlights)
- Hypothesis 1 — IOER differences (3-month, weekly, 2010-2018):
  - IOER differential significant and negative only for Japan in time-series.
  - IOER-differential significantly negative in post-crisis panel and recent-period panel.
  - Quantitative example: ten-basis-point increase of U.S. IOER in 2015W11-2018M6 panel widens 3-month CIP deviations in favor of USD by 2.5 basis points.
  - Selected regression coefficients (panel/aggregates): Δ(r*-r) series include values 0.919***; 1.614***; 0.946***; 0.820***; 1.101***; 1.254***; 1.251***; 1.053***. Δ USDINDEX coefficients include -0.762***; -0.975*; -2.681***; -1.139***; -1.705***; -1.378**; -1.147; -1.435***. Sample sizes: N: 442; 442; 442; 442; 442; 2210; 1350; 860. R-sq range: 0.446; 0.449; 0.316; 0.215; 0.234; 0.333; 0.422; 0.217.
- Hypothesis 2 — Corporate borrowing (5-year, monthly, 2010-2016):
  - Residualized spread significant and positive in panel, consistent with Hypothesis 2.
  - Quantitative example: ten-basis-point reduction in local-currency borrowing costs widens 5-year CIP deviations by one basis point in favor of USD.
  - Selected coefficients: Δ Residualized Spread: 0.0627; 0.00447; 0.0164; 0.0655; 0.0931*; 0.203**; 0.110**. N: 81 per country; Panel N: 405. Panel R-sq: 0.898.
- Hypothesis 3 — MMF holdings (3-month, monthly):
  - Panel regressions: Δ MMF Holdings coefficient statistically significant and positive across full 2011-2018M6 panel.
  - Quantitative example: ten-percent negative shock in prime fund holdings leads to widening of short-term CIP deviations by sixteen basis points in favor of USD (full panel regression).
  - Selected coefficients: Δ MMF Holdings in various regressions: -9.116; -11.21*; 7.395; 7.462; 1.786; 3.305; 1.008; 6.176; 2.187; 4.240; 1.861. Sample sizes: N: 89; 89; 89; 89; 89; 89; 89; 89; 623; 329; 294. R-sq examples: 0.809; 0.700; 0.512; 0.485; 0.315; 0.269; 0.407; 0.688; 0.420; 0.383; 0.470.
  - Note: time-series estimates less precise; panel confirms Hypothesis 3, especially after 2015.
- Hypothesis 4 — Domestic VIX (EUR/GBP, 3-month, 2010-2018M6):
  - EUR: Δ lnVDAX and Δ lnVDAX residual have highly significant and negative coefficients — domestic risk variation explains short-term EUR/USD CIP better than U.S. VIX.
  - GBP: no evidence local-specific risk sentiment materially contributes to wider CIP deviations.
  - Selected examples: Δ lnVDAX: -5.503*; Δ lnVDAX residual: -8.659**. Δ lnVFTSE: 0.328; Δ lnVFTSE residual: 0.237. N: 442; R-sq: 0.320; 0.321; 0.208; 0.208.

### X. Time-varying effects and regime analysis (Euro focus)
- Markov-switching (euro/USD 3-month CIP, 2010-2018M6) — two states:
  - State s=1 (high-volatility episodes) vs s=2 (low-volatility episodes):
    - Δ(r*-r): 1.909*** (s=1); 0.147 (s=2).
    - Δ USDINDEX: -3.352* (s=1); -1.770*** (s=2).
    - Δ lnVIX: -7.528 (s=1); 0.174 (s=2).
    - Δ Forward Bid-Ask: -12.41*** (s=1); -0.199 (s=2).
    - ln(sigma): 1.330*** (s=1); 1.219*** (s=2).
    - P(1 | s): 0.737 (s=1); 0.117 (s=2).
    - N: 441; 441.
  - Interpretation: State 1 corresponds to episodes with stronger volatility and stronger negative co-movement with USDINDEX; State 2 to calmer periods (late 2012–2015).
- Rolling regression (100-week window) for EUR:
  - USDINDEX coefficient most negative during euro crisis period; dollar’s impact rising again recently but remains relatively small.
  - Correlation of rolling USDINDEX coefficients across currencies: high positive correlation among EUR, NOK, SEK; moderate correlation between CHF, GBP, JPY and EUR (selected correlations reported in source).

### XI. Synthesis and policy implications
- Empirical synthesis:
  - CIP violations visible since GFC; can be statistically explained by a set of macrofinancial measures.
  - Consistent key drivers: U.S. dollar strength and FX liquidity conditions; VIX matters to a lesser extent and episodically.
  - Multiple factors (asynchronous monetary policies, prime MMF reforms, regulatory and market-structure changes) contributed; many effects episodic or country-specific.
  - Importance of drivers is time- and country-varying.
- Policy implications:
  - CIP deviations imply financial-market frictions but evidence is insufficient to justify policy intervention in non-crisis times.
  - Further research needed on welfare costs and effects on monetary transmission and exchange-market stability.
  - During crisis times, failure of CIP provides prima facie argument for central bank swap lines to facilitate foreign-currency funding for financial-sector institutions when necessary.

*Source: wp1914 - References ... 26*

### References ... 26

### References ... 26

### I. Introduction — context and motivation
- Covered interest parity (CIP) is defined as the relationship linking the n-period forward premium (f_t,t+n − s_t, annualized, foreign currency per dollar) and the n-period interest-rate differential (r*_t,t+n − r_t,t+n). The n-period CIP deviation x_t,t+n is given by:
  - x_t,t+n = (f_t,t+n − s_t) − (r*_t,t+n − r_t,t+n)
- A “negative dollar basis” is when x_t,t+n < 0, indicating direct dollar funding is cheaper than synthetic dollar funding.
- Historically (pre-GFC) CIP held closely at daily or weekly frequencies; post-GFC persistent CIP deviations have emerged and persisted.
- Policy relevance:
  - CIP deviations may signal financial-market distortions and inefficiencies in resource allocation.
  - Failure of CIP undermines the claim that small open economies can fully insulate domestic monetary policy from U.S. Federal Reserve choices via automatic exchange-rate adjustments.
  - Understanding drivers of CIP departures is key to assessing cross-border monetary policy transmission.

### II. Literature and proposed drivers
- Proposed drivers surveyed include:
  - Regulation-induced or other arbitrage limits (Ivashina, Scharfstein, and Stein 2015; Du, Tepper, and Verdelhan, 2017; Rime, Schrimpf, and Syrstad, 2017).
  - Changes in banks’ balance-sheet capacity tied to U.S. dollar appreciation (Avdjiev et al., 2017).
  - Interest-rate differences across currencies impacting the swap market (Liao, 2016; Brauning and Ivashina, 2017; Sushko et al., 2017).
- Empirical approach:
  - Macro-financial level analysis across ten currencies relative to the U.S. dollar.
  - Methods include time-series and panel estimation, rolling estimation windows, Markov regime-switching models, and split-sample analysis.
  - Alternative funding-rate measures and an instrumental-variable design are used given concerns about Libor as a marginal funding-cost proxy.

### III. Measurement and empirical observations (pre-, during-, and post-GFC)
- Definitions and measures:
  - CIP deviation (cross-currency basis) computed for horizons including 3-month and 5-year.
  - Alternative short-term measure: 3-month U.S. AA-rated financial commercial paper (CP) rate for U.S. dollar borrowing and foreign 3-month government bill rate for foreign lending.
  - FX forward liquidity proxied by forward point bid-ask spreads.
  - Capital-charge measure from Du, Tepper, and Verdelhan (2017) computed from VaR in a 5-year Libor CIP trade.
- Key empirical facts reported:
  - Before the GFC, CIP deviations were very small and fluctuated around zero.
  - During the GFC:
    - Short-term CIP deviations reached levels of about -200 basis points (Figure 1).
    - Five-year horizon deviations were more negative than -50 basis points (Figure 2).
  - Post-GFC:
    - CIP deviations generally decreased toward zero through 2013, widened again after mid-2014.
    - Volatile but persistent CIP deviations appeared during the GFC and have persisted after the GFC (Figure 3).
    - Libor-based and alternative CP–government-bill based bases can differ notably during the GFC, but behavior is somewhat similar after the GFC (Figures 3–4).
  - Currency signs:
    - Most currencies show a negative dollar basis (x_t,t+n < 0), implying direct dollar funding cost advantage if available near Libor.
    - “Carry” currencies such as AUD and NZD display positive deviations (x_t,t+n > 0) relative to the U.S. dollar.
  - FX liquidity:
    - Bid-ask spreads at short-term and long-term maturities are generally small but show spikes with increased frequency during and after the GFC (Figure 5).
  - Capital charges and regulatory cost:
    - Tighter capital requirements and higher capital charges since the GFC appear to have increased the cost of arbitraging (Figure 6).
    - Time-series evolution of capital charges does not fully explain the temporary restoration of CIP from 2013 to 2014 or the subsequent widening after mid-2014.

### IV. Main findings and interpretation
- Three principal conclusions:
  - First, CIP clearly broke down during the GFC and has not held reliably since; CIP deviations increased significantly post-GFC across different benchmarks and currency pairs.
  - Second, while structural factors (e.g., post-crisis financial regulations) likely increased the cost of currency arbitrage, multiple drivers—some temporary—also contributed to variation in CIP deviations over time. Temporary drivers include:
    - Asynchronous monetary policy across the United States, euro area, and Japan.
    - The October 2016 reform of U.S. prime money market funds, which reduced non-U.S. banks’ funding for currency arbitrage and widened CIP deviations.
  - Third, time-series evidence indicates that factors with strong statistical associations across much of the sample (notably U.S. dollar strength) do not have uniform importance across currency pairs and time, suggesting interacting time- and country-specific influences.
- Robust associations across maturities:
  - U.S. dollar strength (Avdjiev et al., 2017), global risk sentiment (proxied by the VIX index), and, to a limited extent, forward exchange-market liquidity (forward point bid-ask spreads) significantly contributed to cross-currency basis variation. These relationships hold across different maturity horizons.

### V. Policy significance and remaining questions
- CIP deviations signal financial-market frictions, but their policy significance (on monetary transmission, exchange-market stability, and welfare costs) is not yet settled.
- Available empirical and theoretical evidence is insufficient to make a definitive case for policy intervention by national or international regulators.
- Further analysis is needed to assess:
  - The magnitude of welfare costs associated with persistent CIP deviations.
  - How deviations alter international monetary policy transmission channels in practice.
  - The relative contributions of slow-moving structural/regulatory factors versus temporary, policy-driven, or market-driven shocks.

*Source: wp1914 - References ... 26*

### 2.2 Basic regression Analysis using Libor-based cross-currency calculations

### 2.2 Basic regression Analysis using Libor-based cross-currency calculations

### Methodology and specification
- Regression approach analogous to uncovered interest rate parity (UIP) literature: regress forward premium (f−s) on interest differential and other determinants.
- Empirical first-difference specification (as presented):
  - *
,,
()
tnt
tt ntt ntt
fs
rr   X
n
αβε
+
++
−
∆=  +  ⋅∆−++δ
- Estimating equation equivalent (after subtracting * ,,( ) tt ntt n rr ++ −∆ ):
  - *
,,
1()()
tnttnttntt
xrr   X
αβε
+++
∆+ −∆+
=−+
δ
- Baseline potential drivers (three):
  - Change in aggregate U.S. dollar strength (FRB U.S. trade-weighted broad dollar index).
  - Log change in the U.S. VIX index (Δ lnVIX) to control for global risk sentiment.
  - Forward bid-ask spreads as a measure of FX market liquidity.
- Baseline data: weekly averages of G8 currencies’ 3-month forward premiums, Libor/interbank rate differential, USDINDEX, and lnVIX from 2002 to 2018 (with sample splits).

### Baseline regression results — Pre-crisis (2002-2006, Table 1a)
- Sample: 2002-2006, weekly, 3-month Libor rate deviations.
- General finding: interest rate differential largely explains changes in forward premium; dollar strength and VIX have little additional power pre-crisis.
- Selected coefficient estimates (Δ(f−s) regressed on Δ(r*−r), Δ USDINDEX, Δ lnVIX, Δ Forward Bid-Ask):
  - Canada (CAD):
    - Δ (r*-r): 0.858*** (0.0347)
    - Δ USDINDEX: -0.360 (0.268)
    - Δ lnVIX: 0.220 (1.673)
    - Δ Forward Bid-Ask: 0.574 (0.822)
    - N: 259, R-sq: 0.779
  - Euro Area (EUR):
    - Δ (r*-r): 0.978*** (0.0288)
    - Δ USDINDEX: -0.0728 (0.157)
    - Δ lnVIX: -1.622 (1.359)
    - Δ Forward Bid-Ask: -0.629** (0.316)
    - N: 259, R-sq: 0.857
  - Panel (CAD CHF DKK EUR GBP JPY NOK SEK):
    - Δ (r*-r): 0.976*** (0.0243)
    - Δ USDINDEX: 0.0687 (0.117)
    - Δ lnVIX: -0.338 (0.421)
    - Δ Forward Bid-Ask: -0.0126 (0.0292)
    - N: 2072, R-sq: 0.798

### Baseline regression results — Crisis (2007-2009, Table 1b)
- Sample: 2007-2009, weekly, 3-month Libor rate deviations.
- General finding: fit worsens relative to pre-crisis (lower R-squared); interest differential tracks forward premium less well; dollar movements matter more.
- Selected coefficient estimates:
  - Canada (CAD):
    - Δ (r*-r): 0.920*** (0.108)
    - Δ USDINDEX: -0.956 (1.504)
    - Δ lnVIX: 8.125 (13.04)
    - Δ Forward Bid-Ask: -2.426 (2.069)
    - N: 156, R-sq: 0.617
  - Euro Area (EUR):
    - Δ (r*-r): 0.820*** (0.150)
    - Δ USDINDEX: 1.246 (2.629)
    - Δ lnVIX: 5.873 (13.22)
    - Δ Forward Bid-Ask: -10.38*** (3.821)
    - N: 156, R-sq: 0.563
  - Panel:
    - Δ (r*-r): 0.891*** (0.0274)
    - Δ USDINDEX: -1.327*** (0.308)
    - Δ lnVIX: 0.0815 (2.753)
    - Δ Forward Bid-Ask: -0.473 (0.273)
    - N: 1248, R-sq: 0.481

### Baseline regression results — Post-crisis (2010-2018M6, Table 1c)
- Sample: 2010-2018M6, weekly, 3-month Libor rate deviations.
- General finding: significance of dollar strength and VIX rises; USDINDEX and lnVIX significant in panel and many time-series regressions; FX liquidity significant for CHF, EUR, and panel.
- Selected coefficient estimates:
  - Switzerland (CHF):
    - Δ (r*-r): 1.572*** (0.207)
    - Δ USDINDEX: -0.967* (0.573)
    - Δ lnVIX: -6.424** (3.127)
    - Δ Forward Bid-Ask: -3.342*** (0.709)
    - N: 442, R-sq: 0.448
  - Japan (JPY):
    - Δ (r*-r): 0.898*** (0.240)
    - Δ USDINDEX: -1.735*** (0.427)
    - Δ lnVIX: -5.846** (2.415)
    - Δ Forward Bid-Ask: -1.394 (0.858)
    - N: 442, R-sq: 0.198
  - Panel:
    - Δ (r*-r): 1.082*** (0.186)
    - Δ USDINDEX: -1.432*** (0.248)
    - Δ lnVIX: -3.556** (1.357)
    - Δ Forward Bid-Ask: -0.0557* (0.0264)
    - N: 3536, R-sq: 0.331

### Regression analysis using alternative marginal funding costs and IV approach
- Motivation: Libor-based measures may not track actual marginal funding costs for various market participants.
- Two approaches:
  1. Replace Libor-based differential with 3-month foreign government bill rate − US commercial paper differential (representative of a 3-month U.S. AA-rated financial commercial paper issuer marginal funding cost) in OLS regressions.
  2. Use the commercial paper − treasury bill differential as an instrument for the Libor-based rate differential in time-series and panel IV regressions.
- Rationale: if both differentials are noisy proxies with classical measurement error, IV provides consistent coefficient estimates.

### IV regression results (Commercial Paper-Treasury Bill rate as instrument)
- Tables 2a–2c summarize IV estimates across pre-crisis, crisis, and post-crisis periods. Broad conclusions:
  - Similar qualitative conclusions to OLS: dollar strength, VIX, and liquidity significant in post-crisis samples; rarely significant before and during the crisis.
  - During crisis, panel estimate of USD strength flips sign and becomes significantly positive in one specification—possibly reflecting divergence between Libor and commercial paper bases in late 2008.
- Selected IV coefficient estimates — Pre-crisis (Table 2a):
  - Panel (CAD EUR JPY NOK SEK):
    - Δ (r*-r): 0.980*** (0.0211)
    - Δ USDINDEX: -0.0568 (0.0930)
    - Δ lnVIX: 0.581 (0.841)
    - Δ Forward Bid-Ask: 0.0154 (0.0406)
    - N: 1295, R-sq: 0.814
- Selected IV coefficient estimates — Crisis (Table 2b):
  - Panel:
    - Δ (r*-r): 1.517*** (0.133)
    - Δ USDINDEX: 1.759*** (0.583)
    - Δ lnVIX: -5.056 (5.639)
    - Δ Forward Bid-Ask: -0.367* (0.189)
    - N: 755, R-sq: 0.651
- Selected IV coefficient estimates — Post-crisis (Table 2c):
  - Panel:
    - Δ (r*-r): 1.101*** (0.265)
    - Δ USDINDEX: -1.667*** (0.314)
    - Δ lnVIX: -2.866*** (0.902)
    - Δ Forward Bid-Ask: -0.0282** (0.0121)
    - N: 2192, R-sq: 0.408

### Key empirical implications and interpretation
- Pre-crisis (2002-2006): changes in interest rate differentials largely explained variations in forward premiums; dollar strength and risk sentiment had limited additional explanatory power.
- Crisis period (2007-2009): the explanatory power of interest differentials weakened (lower R-squared); dollar strength and liquidity effects became more important in panel estimates; large idiosyncratic and measurement divergences observed across funding-rate proxies.
- Post-crisis (2010-2018M6): dollar strength (Δ USDINDEX) and risk sentiment (Δ lnVIX) gained prominence as determinants of CIP deviations across panel and many individual currency regressions; FX liquidity (Δ Forward Bid-Ask) also matters for specific currencies (CHF, EUR) and in panel regressions.
- Using alternative marginal funding cost measures and IV estimation:
  - Broad qualitative results are similar, supporting robustness of the role of USD strength, VIX, and liquidity in the post-crisis period.
  - Divergences between Libor-based and commercial paper–treasury bill bases during crisis episodes (notably late 2008) can affect coefficient signs and magnitudes, indicating important measurement and market-structure effects during stress periods.

*Source: wp1914 - 2.2 Basic regression Analysis using Libor-based cross-currency calculations (IMF working paper PDF).*

### 3.1 Review of literature and summary of additional hypotheses

### 3.1 Review of literature and summary of additional hypotheses

### Monetary policy divergence
- Empirical observation:
  - Figure 7 (reproduced and extended from Du, Tepper and Verdelhan (2017)) plots a highly positive cross-sectional relationship between G10 countries’ period-average interest rates (measured by Libor) and their period-average cross-currency bases against the USD for 2010-18 and 2015-18, respectively.
  - Countries with lower interest rate tend to exhibit more negative cross-currency dollar bases. This relationship seems to be even stronger during the recent period (the right panel in Figure 7).
- Mechanisms from literature:
  - Brauning and Ivashina (2017): domestic monetary easing widens the difference between foreign and domestic interest rates, leading global banks to increasingly borrow from local-currency deposit facilities; this raises currency hedging costs and contributes to widening cross-currency dollar bases.
  - Recent policy divergence (example: ECB quantitative easing amid U.S. monetary tightening) may produce favorable borrowing conditions in non-USD currencies. Lower borrowing cost in euros relative to the U.S. dollar could encourage corporate bond issuance in euros (Liao (2016)); conversion of proceeds into USD raises the price of dollar swaps and widens the basis.
- Data and controls used in this study:
  - Use both Liao’s (2016) series on residualized credit spread (variable Residualized_Spread) and IOER differential as proxies (IOER includes interest rate on central banks’ deposit facilities, following Brauning and Ivashina (2017)).
  - Additional controls: dollar strength (change in the USD index), risk sentiment (change in log VIX), and changes in bid-ask spreads.
- Hypotheses:
  - Hypothesis 1 (IOER differential). Controlling for other factors, a larger difference between the deposit facility rate in the central bank where a foreign global bank is headquartered and U.S. IOER (IOER* - IOER_us) leads to a decrease in the FX hedge demand (which would reduce a cross-country dollar negative basis, implying a negative regression coefficient in our framework).
  - Hypothesis 2 (Corporate Issuance). Controlling for other factors, the more favorable borrowing conditions are in domestic currencies compared with those in USD (the lower the variable Residualized_Spread), the higher would be the demand of currency hedges, implying a widening of cross-country dollar negative basis, and a positive regression coefficient in our framework).

### Regulatory constraints and dollar funding sources
- Literature links CIP deviations to limits to arbitrage due to post-GFC regulation:
  - Du, Tepper, and Verdelhan (2017) and Rime, Schrimpf, and Syrstad (2017) discuss effects of banking regulatory instruments that increase costs of currency arbitrage.
  - Brauning and Puria (2017) find that higher bank balance sheet costs, together with increasing USD demand due to monetary policy divergence, push up the price of dollar swaps and amplify CIP deviations.
  - Low-frequency changes in prudential regulatory instruments (e.g., capital requirements) are unlikely to explain high-frequency movement in the cross-currency basis.
- Prime MMF reform channel:
  - Foreign bank branches in the U.S. rely on non-bank short-term dollar funding mainly via commercial paper and certificates of deposit held by U.S. prime money market funds (MMF).
  - Prime money market fund reform in late 2015 triggered large outflows from prime MMFs to government MMFs; foreign banks lost considerable dollar funding, potentially resulting in more deeply negative cross-currency bases (Iida, Kimura, and Sudo (2016); Du, Tepper, and Verdelhan (2017); Nakaso (2017)).
- Data construction and role in regressions:
  - Monthly gross holdings data for U.S. prime money market funds obtained from the Office of Financial Research Money Market Fund Monitor.
  - For each month t and each currency i, variable itHoldings is constructed by assigning the aggregate value of holdings of U.S. MMFs on bank branches headquartered in the country for which currency i is official (example: ,JPY_tHoldings includes prime MMFs’ claims on all Japanese bank branches in the United States; ,EUR_tHoldings sums holdings on branches of banks headquartered in France, Germany, Netherlands, Belgium, Austria, Spain, Italy, and Luxembourg).
  - itHoldings is included in baseline time-series and panel regressions along with dollar strength and the VIX.
- Hypothesis:
  - Hypothesis 3 (Dollar funding strain). Controlling for other factors, a negative shock to prime money market fund holdings (a lower value of Holdings) leads to a higher CIP deviation in favor of USD (a widening of cross-country dollar negative bases, implying a positive regression coefficient).

### Domestic risk sentiment
- Empirical motivation:
  - Table 1c suggests U.S. VIX may drive CIP deviations for a number of individual currencies but does not work well for the euro and the pound sterling.
- Approach:
  - Use U.S. VIX counterparts in Europe: VDAX-New Index for Germany and VFTSE Index for the UK to test domestic risk effects on corresponding currencies’ CIP deviations versus USD.
- Hypothesis:
  - Hypothesis 4 (Domestic risk prospect). Controlling for other factors, a rise in the VDAX index widens CIP deviations of EUR/USD in favor of USD. Similarly, a rise in the VFTSE index widens CIP deviations of GBP/USD in favor of USD (implying negative coefficients for both indices).

*Source: wp1914 - 3.1 Review of literature and summary of additional hypotheses*

### 3.2 Empirical tests: Results

### 3.2 Empirical tests: Results

### Hypothesis 1 — IOER Differences and CIP Deviations (3-month, weekly regression)
- Sample and setup:
  - Time-series regressions: 2010-2018, individual-currency.
  - Panel regressions: sample split by start of ECB’s QE program (March 2015), panels (10M1-18M6; 10M1-15W10; 15W11-18M6).
  - Additional control: weekly IOER differential (Δ (IOER* - IOER)).
- Main findings:
  - The coefficient of the IOER differential in time-series regressions is statistically significant and negative only for Japan.
  - The IOER-differential coefficient is significantly negative in the post-crisis (2010-18M6) panel and the recent-period (2015W11-2018M6) panel.
  - IOER differentials grew substantially after 2015, indicating monetary policy divergence (U.S. tightening vs. sustained easing by other major central banks).
  - Quantitative example: A ten-basis-point increase of the U.S. IOER in the 2015W11-2018M6 panel regression, holding foreign IOER unchanged, would widen 3-month CIP deviations in favor of the U.S. dollar by 2.5 basis points.
- Selected coefficient estimates (Δ(f-s) regressions):
  - Δ (r*-r): 0.919***; 1.614***; 0.946***; 0.820***; 1.101***; 1.254***; 1.251***; 1.053*** (standard errors in parentheses)
  - Δ USDINDEX: -0.762***; -0.975*; -2.681***; -1.139***; -1.705***; -1.378**; -1.147; -1.435*** 
  - Δ lnVIX: -0.642; -6.322**; -1.395; 0.176; -5.697**; -3.079*; -7.502**; 0.218
  - Δ Forward Bid-Ask: -0.434; -3.412***; -4.981**; -0.277; -1.343; -2.515*; -2.754*; -1.796*
  - Δ (IOER* - IOER): -0.0999; -0.118; -0.135; -0.111; -0.394**; -0.211***; 0.155*; -0.253***
- Sample sizes and fit:
  - N: 442; 442; 442; 442; 442; 2210; 1350; 860
  - R-sq: 0.446; 0.449; 0.316; 0.215; 0.234; 0.333; 0.422; 0.217

### Hypothesis 2 — Corporate Borrowing Conditions and CIP Deviations (5-year, monthly regression, 2010-2016)
- Data and setup:
  - Monthly data: 2010-2016.
  - Use 5-year FX forwards and 5-year interest rates for long-term forward premiums and interest rate differentials.
- Main findings:
  - Time-series regression coefficient for change in the credit spread is significant and positive only for Japan.
  - Panel regression coefficient for residualized spread is significantly positive, consistent with Hypothesis 2.
  - Quantitative example: A ten-basis-point reduction in local-currency borrowing costs widens 5-year CIP deviations by one basis point in favor of the U.S. dollar.
  - Dollar strength effect is visible; VIX and FX liquidity effects are weaker.
- Selected coefficient estimates (Δ(f-s) regressions):
  - Δ (r*-r): 0.876***; 0.971***; 0.954***; 1.141***; 1.124***; 0.920***; 0.995***
  - Δ USDINDEX: 0.0559; -0.420; -1.690***; -2.240***; -0.582**; -0.825; -1.228**
  - Δ lnVIX: -2.366; 2.633**; -5.183; 2.164; -5.660; 1.141; -0.434
  - Δ Forward Bid-Ask: -0.0188; 0.0405; -0.0199; -0.172**; 0.0231**; -0.140; 0.0123
  - Δ Residualized Spread: 0.0627; 0.00447; 0.0164; 0.0655; 0.0931*; 0.203**; 0.110**
- Sample sizes and fit:
  - N: 81; 81; 81; 81; 81; 81; 405
  - R-sq: 0.955; 0.961; 0.909; 0.909; 0.920; 0.890; 0.898

### Hypothesis 3 — Money Market Fund Holdings and CIP Deviations (3-month, monthly regression)
- Data and setup:
  - Monthly frequency; change in log gross prime MMF holdings added as control; coefficient interpretable as a semi-elasticity.
  - OFR MMF monitor data compiled from SEC Form N-MFP filings since 2011.
- Main findings:
  - Time-series regressions yield less precise estimates; panel regressions largely confirm Hypothesis 3.
  - Coefficient on changes in MMF holdings is statistically significant and positive across the full 2011-2018M6 panel sample.
  - Effect is especially strong after 2015, coinciding with proposal to require floating NAV for prime MMFs and substantial prime MMF outflows.
  - Quantitative example: In the full panel regression, a ten-percent negative shock in prime fund holdings leads to the widening of short-term CIP deviations by sixteen basis points in favor of the dollar.
  - Dollar appreciation (Δ USDINDEX) and worsening FX liquidity (Δ Forward Bid-Ask) raise the cost of synthetic dollar borrowing relative to direct dollar funding.
  - VIX results are less robust and flip sign in the most recent period (2015M1-2018M6).
- Selected coefficient estimates (Δ(f-s) regressions):
  - Δ (r*-r): 0.925***; 0.935***; 1.236***; 0.718***; 0.645***; 0.654***; 0.560***; 1.052***; 0.873***; 0.840***; 0.926***
  - Δ USDINDEX: 0.120; -0.664; -0.963; -2.500***; -1.133*; -1.680**; -2.264***; -1.458***; -1.702***; -1.424; -1.862***
  - Δ lnVIX: -2.666; -1.385; -11.94; -8.427; -0.905; -8.109; -4.910; -0.987; -7.070*; -19.67**; 3.182**
  - Δ Forward Bid-Ask: -3.242; -0.737; -8.281**; -15.53**; -0.122; -2.106; -0.179; -0.0795; -0.161**; -0.102; -0.142***
  - Δ MMF Holdings: -9.116; -11.21*; 7.395; 7.462; 1.786; 3.305; 1.008; 6.176; 2.187; 4.240; 1.861
- Sample sizes and fit:
  - N: 89; 89; 89; 89; 89; 89; 89; 89; 623; 329; 294
  - R-sq: 0.809; 0.700; 0.512; 0.485; 0.315; 0.269; 0.407; 0.688; 0.420; 0.383; 0.470

### Hypothesis 4 — Domestic VIX and CIP Deviations (3-month, 2010-2018M6)
- Setup:
  - Time-series regressions for EUR and GBP, replacing changes in log U.S. VIX with domestic implied volatility indices (VDAX-New for euro area, VFTSE for UK).
  - Also residualize domestic volatility by regressing log levels of VDAX-New and VFTSE on log U.S. VIX and use residuals.
- Main findings:
  - For the euro (EUR), domestic variation in implied volatility explains short-term CIP deviations better than the U.S. VIX:
    - Changes in both log level of VDAX-New and residualized VDAX-New have highly significant and negative coefficients.
    - Interpretation: A surge in euro area-wide risk sentiment increases attractiveness of dollar-denominated investment, boosting demand for dollar swaps from euros and widening CIP deviations in favor of the U.S. dollar.
  - For the United Kingdom (GBP), no evidence that local-specific risk sentiment materially contributes to wider CIP deviations.
- Selected coefficient examples (EUR/GBP regressions):
  - Δ (r*-r): 0.871***; 0.866***; 0.734***; 0.735***
  - Δ USDINDEX: -2.527***; -2.825***; -1.163***; -1.144***
  - Δ Forward Bid-Ask: -4.649*; -4.833**; -0.178; -0.178
  - Δ lnVDAX: -5.503*
  - Δ lnVDAX residual: -8.659**
  - Δ lnVFTSE: 0.328
  - Δ lnVFTSE residual: 0.237
- Sample sizes and fit:
  - N: 442; 442; 442; 442
  - R-sq: 0.320; 0.321; 0.208; 0.208

### Time-varying explanatory power of aggregate U.S. dollar strength (Euro focus)
- Methods:
  - Dynamic two-state Markov-switching time-series regression for euro/USD 3-month CIP deviations (2010-2018M6).
  - Alternative: 100-week window rolling regression to evaluate time-varying USDINDEX coefficient.
- Markov-switching results (Table 7):
  - State s = 1 vs s = 2 coefficient estimates:
    - Δ (r*-r): 1.909*** (s=1); 0.147 (s=2)
    - Δ USDINDEX: -3.352* (s=1); -1.770*** (s=2)
    - Δ lnVIX: -7.528 (s=1); 0.174 (s=2)
    - Δ Forward Bid-Ask: -12.41*** (s=1); -0.199 (s=2)
  - ln(sigma): 1.330*** (s=1); 1.219*** (s=2)
  - P(1 | s): 0.737 (s=1); 0.117 (s=2)
  - N: 441; 441
- Interpretation:
  - Both states are highly persistent.
  - State 1 is associated with episodes of stronger volatility in the cross-currency basis and stronger negative co-movement with dollar strength (e.g., euro area crisis and period after ECB QE).
  - State 2 corresponds to periods with smaller, less volatile CIP deviations (e.g., late 2012–2015).
  - The USDINDEX coefficient is substantially more negative in state 1 than in state 2.
- Rolling regression findings:
  - The 100-week rolling USDINDEX coefficient for EUR shows the most negative (largest impact) during the euro crisis period; the dollar’s impact has been rising again recently but remains relatively small.
  - Correlation of rolling USDINDEX coefficients across currencies (Table 8) shows:
    - High positive correlation among EUR, NOK, SEK.
    - Moderate positive correlation between CHF, GBP, JPY and EUR.
  - Table 8 correlation matrix (3-month basis, 2010-2017) entries (selected):
    - AUD–AUD: 1.000
    - CAD–EUR: 0.759
    - EUR–NOK: 0.846
    - NOK–SEK: 0.962
    - JPY–EUR: 0.586
    - (Full matrix rows/columns preserved in source)

### Conclusions and policy implications
- Empirical synthesis:
  - CIP violations are visible across cross-currency dollar basis since the GFC and can be statistically explained by a set of macrofinancial measures available before, during, and after the GFC.
  - Key drivers with consistent explanatory power: U.S. dollar strength and FX liquidity conditions; the VIX matters to a lesser extent and is more relevant in some recent periods.
  - Multiple factors (asynchronous monetary policies, prime MMF reforms, regulatory and market-structure changes) contributed to CIP deviations; many effects are episodic or country-specific.
  - Time-series evidence shows that the importance of drivers such as U.S. dollar strength is not uniform across currencies or time, indicating interactions among several time- and country-specific factors.
- Policy implications:
  - CIP deviations imply financial-market frictions but evidence is insufficient to justify policy intervention in non-crisis times.
  - More research is needed to assess whether frictions significantly undermine monetary policy transmission or exchange-market stability.
  - During crisis times, failure of CIP provides a prima facie argument for the importance of central bank swap lines to facilitate foreign-currency funding for financial-sector institutions when necessary.

*Source: wp1914 - 3.2 Empirical tests: Results*

### REFERENCES

### wp1914 - REFERENCES

### References
- Full list of cited works includes (selection from source):
  - Akram, Q. Farooq, Dagfinn Rime, and Lucio Sarno. 2008. "Arbitrage in the foreign exchange market: Turning on the microscope." Journal of International Economics 76 (2): 237-253.
  - Amador, Manuel, Javier Bianchi, Bocola, Luigi, and Fabrizio Perri. 2017. "Exchange rate policies at the zero lower bound." NBER Working Paper.
  - Arsov, Ivailo, Greg Moran, Ben Shanahan, and Karl Stacey. 2013. "OTC Derivatives Reforms and the Australian Cross-currency Swap Market." Reserve Bank of Australia Bulletin.
  - Avdjiev, Stefan, Wenxin Du, Catherine Koch, and Hyun Song Shin. 2017. "The dollar, bank leverage and the deviation from covered interest parity." BIS Working Paper.
  - Baba, N., and F. Packer. 2009. "Interpreting Deviations from Covered Interest Parity during the Financial Market Turmoil of 2007–08." Journal of Banking and Finance 33 (11): 1953-62.
  - Brauning, Falk, and Kovid Puria. 2017. "Uncovering covered interest parity: The rold of bank regulation and monetary policy." Federal Reserve Bank of Boston Current Policy Perspectives.
  - Brauning, Falk, and Victoria Ivashina. 2017. "Monetary policy and global banking." NBER Working Paper.
  - Callaghan, Michael. 2017. The New Zealand Dollar in Global Markets. Reserve Bank of New Zealand Bulletin 80 (November): 3-17.
  - Cecchetti, Stephen, and Kermit Schoenholtz. 2017. "Eclipsing LIBOR." Commentary.
  - Cenedese, Gino, Pasquale Della Corte, and Tianyu Wang. 2017. "Currency Mispricing and Dealer Balance Sheets." Working Paper.
  - Du, Wenxin, Alexander Tepper, and Adrien Verdelhan. 2017. "Deviations from Covered Interest Rate Parity." NBER Working Paper.
  - Duffie, Darrell, and Jeremy C. Stein. 2015. "Reforming LIBOR and Other Financial Market Benchmarks." Journal of Economic Perspective 29 (2): 191-212.
  - Fama, Eugene F. 1984. "Forward and spot exchange rates." Journal of Monetary Economics 14: 319-338.
  - Fukuda, Shin-ichi, and Mariko Tanaka. 2017. "Monetary policy and covered interest parity in the post GFC period: Evidence from the Australian dollar and the NZ dollar." Journal of International Money and Finance 301-317.
  - ICE Benchmark Administration. 2016. "Roadmap to ICE Libor."
  - Iida, Tomoyuki, Takeshi Kimura, and Nao Sudo. 2018. "Regulatory reforms and the dollar funding of global banks: Evidence from the impact of monetary policy divergence." Bank of Japan Working Paper.
  - Ivashina, V., D.S. Sharfstein, and J.C. Stein. 2015. "Dollar Funding and the Lending Behavior of Global Banks." Quarterly Journal of Economics 130 (3): 1241-81.
  - Keynes, John M. 1923. A Tract on Monetary Reform. London: Macmillan.
  - Liao, Gordon. 2016. "Credit migration and covered interest rate parity." Working Paper.
  - Nakaso, Hiroshi. 2017. "Monetary policy divergence and global financial stability: From the perspective of demand and supply of safe assets." Speech.
  - Pinnington, James, and Maral Shamloo. 2016. "Limits to arbitrage and deviations from covered interest rate parity." Bank of Canada Staff Discussion Paper.
  - Rime, Dagfinn, Andreas Schrimpf, and Olav Syrstad. 2017. "Segmented money markets and covered interest parity arbitrage." BIS Working Paper.
  - Sushko, Vladyslav, Claudio Borio, Robert McCauley, and Patrick McGuire. 2017. "The failure of covered interest parity: FX hedging demand and costly balance sheets." BIS Working Paper.
  - Taylor, Mark. 1989. “Covered interest arbitrage and market turbulence.” Economic Journal 99, 376–391.

### Appendix: Additional Tables — Variable Notation and Data Sources (Table A1)
- Forward Premium f-s: Bloomberg. For five-year frequency, use 5-year forward rate.
- Libor Rate Differential r*-r: Haver Analytics, Bloomberg. Asterisk denotes foreign (as opposed to U.S.). For five-year frequency, use 5-year interest rate swap.
- Commercial Paper - Treasury Bill Rate Differential r*TB - rCP: Bloomberg. Asterisk denotes foreign (as opposed to U.S.).
- Trade Weighted U.S. Dollar Index: Broad USDINDEX: Haver Analytics.
- Forward Point Bid-Ask Spread Forward Bid-Ask: Bloomberg. In unit of forward point difference.
- U.S. VIX Index VIX: Haver Analytics.
- FTSE 100 VIX Index VFTSE: Bloomberg.
- Dax-New Volatility Index VDAX: Bloomberg.
- Interest Differential on Excess Reserves IOER*-IOER: Haver Analytics, Brauning and Ivashina (2017). Asterisk denotes foreign (as opposed to U.S.). Definition of IOER comes from Brauning and Ivashina (2017), including deposit facility rates.
- Currency-specific corporate credit spread against USD Residualized Spread: Liao (2016).
- Prime Money Market Holdings MMF Holdings: Office of Financial Research. In log billions USD. Each currency is mapped to its country/region, which in turn is mapped to the ultimate parent country identified by OFR.

### Table A2: Baseline Regressions, Commercial Paper-Treasury Bill Rate Differences (3-month horizon) — Panel coefficients (selected)
- Panel (Canada, Euro Area, Japan, Norway, New Zealand, Sweden, Panel)
  - Δ (r*-r): 0.725*** (Canada), 0.704*** (Euro Area), 0.836*** (Japan), 0.350*** (Norway), 0.586*** (New Zealand), 0.751*** (Sweden), 0.545*** (Panel)
    - Standard errors: (0.0474), (0.0490), (0.0569), (0.0837), (0.0644), (0.0498), (0.0994)
  - Δ USDINDEX: -0.638* (Canada), 0.213 (Euro Area), -0.119 (Japan), 0.308 (Norway), 1.003** (New Zealand), 0.191 (Sweden), -0.0286 (Panel)
    - Standard errors: (0.352), (0.278), (0.313), (0.611), (0.492), (0.335), (0.240)
  - Δ lnVIX: 0.435 (Canada), -0.925 (Euro Area), 4.174 (Japan), 9.345* (Norway), 1.665 (New Zealand), -0.665 (Sweden), 2.135 (Panel)
    - Standard errors: (2.030), (2.057), (2.545), (5.618), (4.000), (2.759), (1.484)
  - Δ Forward Bid-Ask: 1.298 (Canada), -1.550*** (Euro Area), 0.731 (Japan), 0.0903 (Norway), 0.904 (New Zealand), -0.0204 (Sweden), 0.0337 (Panel)
    - Standard errors: (0.990), (0.413), (0.794), (0.0977), (0.986), (0.0562), (0.0439)
  - N: 259 (Canada), 259 (Euro Area), 259 (Japan), 259 (Norway), 254 (New Zealand), 259 (Sweden), 1549 (Panel)
  - R-sq: 0.627 (Canada), 0.611 (Euro Area), 0.543 (Japan), 0.219 (Norway), 0.409 (New Zealand), 0.569 (Sweden), 0.376 (Panel)

- (a) Pre-crisis (2002-2006)
  - Δ (r*-r): 0.248*** (Canada), 0.417*** (Euro Area), 0.520* (Japan), 0.445*** (Norway), 0.276*** (New Zealand), 0.472*** (Sweden), 0.408*** (Panel)
    - Standard errors: (0.0886), (0.118), (0.264), (0.125), (0.0813), (0.158), (0.0569)
  - N: 151 (each country), 906 (Panel). R-sq (Panel): 0.273 (Panel).

- (b) Crisis (2007-2009)
  - Δ (r*-r): 0.302*** (Canada), 0.378*** (Euro Area), 0.0804 (Japan), 0.177*** (Norway), 0.256*** (New Zealand), 0.159** (Sweden), 0.227*** (Panel)
    - Standard errors: (0.0660), (0.0662), (0.0665), (0.0410), (0.0527), (0.0715), (0.0423)
  - Δ USDINDEX: -1.245*** (Canada), -2.475*** (Euro Area), -2.038*** (Japan), -1.965*** (Norway), -0.516 (New Zealand), -2.599*** (Sweden), -1.944*** (Panel)
    - Standard errors: (0.282), (0.399), (0.449), (0.454), (0.387), (0.570), (0.287)
  - N: 442 (most countries), 2480 (Panel). R-sq (Panel): 0.156.

- (c) Post-crisis (2010-2018M6)
  - Δ (r*-r): 0.894*** (Canada), 0.628*** (Euro Area), 0.880*** (Japan), 0.854*** (Norway), 0.819*** (New Zealand), 0.789*** (Sweden), 0.979*** (Panel countries subset), 0.740*** (Sweden in extended panel), 0.850*** (Panel)
    - Standard errors listed in table.
  - N totals and R-sq vary by country; example Panel N: 1799, R-sq: 0.545.

### Table A3: Baseline Regressions (5-year horizon) — Selected coefficients
- Panel (Canada, Switzerland, Denmark, Euro Area, UK, Japan, Norway, Sweden, Panel)
  - Δ (r*-r): 0.867*** (Australia), 0.982*** (Canada), 0.927*** (Switzerland), 0.952*** (Denmark), 0.997*** (Euro Area), 0.966*** (UK), 0.835*** (Japan), 0.994*** (Norway), 0.847*** (New Zealand), 0.971*** (Sweden), 0.947*** (Panel)
    - Example standard errors: (0.0270), (0.0301), (0.0309), (0.0455), (0.0319), (0.0307), (0.0396), (0.0374), (0.0191), (0.0314), (0.0188)
  - Δ USDINDEX coefficients across countries include: 0.937 (Australia), -0.715* (Canada), -0.582 (Switzerland), -2.756** (Denmark), -1.067* (Euro Area), -0.865 (UK), -0.508 (Japan), -0.908 (Norway), 0.0483 (New Zealand), -0.133 (Sweden), -1.144*** (Panel)
  - Δ lnVIX: significant negative coefficients for several countries (example: -12.77** for Australia, -21.94** for Denmark, -12.87** for Euro Area, -17.90*** for UK, -13.00*** for Panel)
  - Interaction terms with Capital Charge and Bid-Ask reported with coefficients and standard errors.
  - N: 53 (Australia), 55 (Canada), 55 (Switzerland), 50 (Denmark), 55 (Euro Area), 55 (UK), 53 (Japan), 53 (Norway), 55 (New Zealand), 51 (Sweden), Panel N: 427.
  - R-sq: ranges by country; example Panel R-sq: 0.938.

- Table A4: Interaction with Du, Tepper and Verdelhan (2017) capital charges (2002-2015, 5-year horizon, quarterly regression)
  - Reports Δ USDINDEX * Capital Charge, Δ lnVIX * Capital Charge, Δ Bid-Ask * Capital Charge coefficients and standard errors for each country. Example:
    - Δ lnVIX * Capital Charge for Sweden: 2.871** (standard error (0.905))
  - N: 427. Panel R-sq: 0.938.

### Figure A5
- Figure A5: Estimated USDINDEX coefficient from 100-week rolling regression (3-month CIP deviations).

*Italic: Source: wp1914 - REFERENCES (source PDF filename: wp1914 - REFERENCES).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wp1914.pdf_