💰 quantitative finance

A Three-Variable Benchmark for Post-GFC Covered Interest Parity Deviations

This paper introduces a robust daily-frequency benchmark for post-GFC covered interest parity deviations using three lagged public state variables (NFCI, the U.S. dollar index, and the Treasury yield curve slope) that effectively captures persistent background components across G10 and KRW currency-tenor panels.

Original authors: Useong Shin

Published 2026-05-26

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Useong Shin

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the global financial system as a massive, high-stakes game of "exchange rate poker." In a perfect world, the rules of Covered Interest Parity (CIP) would mean that no matter which currency you play with, the cost of borrowing money in one currency and swapping it for another should be exactly the same. There should be no "free money" or easy tricks to make a profit just by switching currencies.

However, since the 2008 Global Financial Crisis (GFC), the game has gotten messy. The rules aren't holding up perfectly. Sometimes, the cost to swap currencies is higher or lower than it "should" be. These gaps are called CIP deviations.

For years, economists have been trying to figure out why these gaps exist. They've found that things like bank regulations, how much cash banks have in their pockets, and how much investors want to hedge their bets all play a part. But there was a problem: there was no simple, standard "ruler" to measure these daily gaps. It was like trying to judge a new car's speed without a standard speedometer; everyone was just making up their own rules.

Enter this paper's solution: The "Three-Variable Benchmark."

The author, Useong Shin, asks a simple question: Can we explain these messy currency gaps using just three simple, public numbers that anyone can look up?

He proposes a "recipe" using three ingredients:

The NFCI (National Financial Conditions Index): Think of this as the weather report for the financial world. Is it stormy and tight (hard to get money)? Or is it sunny and loose?
The Broad Dollar Index: This is the global tide. Is the US dollar rising like a high tide, making everything else feel lower? Or is it receding?
The Treasury Slope: This looks at the difference between long-term and short-term US interest rates. It's like checking the shape of the hill the economy is climbing. Is the hill steep (good growth expectations) or flat (sluggish growth)?

The Results: A Surprisingly Good Fit

The author tested this "three-ingredient recipe" against daily data from 2008 to 2025, covering major currencies like the Euro, Yen, and Korean Won.

Here is what he found, using simple analogies:

It's a Strong Map, Not a GPS: The recipe explains about 64% of the daily ups and downs in these currency gaps. It doesn't predict every single tiny movement (like a GPS that tells you exactly where a pothole is), but it is excellent at describing the general terrain. It captures the "background noise" of the market.
It Works Even When You Hide Data: The author tried a trick called "Leave-One-Year-Out." He trained the model on data from 2008–2024, then hid 2025 to see if it could guess what happened. It did a great job. This proves the recipe isn't just memorizing the past; it's actually understanding the underlying forces.
It's Not Just About "Quarter-End" Spikes: Some previous studies said currency gaps get crazy only at the very end of a quarter (when banks are checking their books). This paper says: "Not really." Because the data starts at 3-month contracts (not 1-week ones), this recipe explains the steady, long-term drift of the market, not just the frantic last-minute scrambling.
It's Not Just "Risk" (VIX): The author checked if this was just a fancy way of measuring general fear (the VIX index). It's not. While fear matters, these three specific variables (Financial Conditions, Dollar Strength, and Yield Curve Shape) tell a unique story that fear alone cannot.

What This Means for You (The General Audience)

Think of this paper as creating a standardized "Thermometer" for the global currency market.

Before this, if a researcher wanted to say, "My new theory about bank regulations explains currency gaps!" they could compare it to a weak, made-up model and look like a genius.

Now, they have to compare their new theory to this Three-Variable Benchmark.

If their new theory can't beat this simple recipe, it probably isn't adding much value.
If their new theory does beat it, then they have genuinely found something new and important.

The Bottom Line

The paper doesn't claim to have solved the mystery of why currency gaps exist forever. Instead, it provides a simple, public, and reliable baseline.

It tells us that the messy, daily gaps in currency prices are largely driven by three big, observable forces: how tight money is, how strong the dollar is, and what the US interest rate curve looks like.

It's a reminder that even in a complex, high-tech financial world, sometimes the biggest trends can be understood by looking at just three simple, public numbers.

Technical Summary: A Three-Variable Benchmark for Post-GFC Covered Interest Parity Deviations

Problem Statement
Covered Interest Parity (CIP) is a fundamental no-arbitrage condition in international finance, yet post-Global Financial Crisis (GFC) data reveals persistent deviations. While daily CIP deviation panels are now available (e.g., Du et al., 2025b), the empirical literature lacks a canonical, public-data benchmark for daily regressions comparable to standard factor models in asset pricing. Existing determinants of CIP deviations—such as intermediary balance-sheet constraints, regulatory frictions, and liquidity measures—are often low-frequency, proprietary, or difficult to reconstruct daily. Consequently, new variables are frequently judged against ad hoc specifications rather than a transparent, reproducible standard. This paper addresses this gap by asking whether a small set of public macrofinancial state variables can summarize the persistent component of government-bond CIP deviations sufficiently to serve as a daily benchmark.

Methodology
The study utilizes government-bond CIP deviation data ( $x^{Govt}_{i,n,t}$ ) from Du et al. (2025b), covering G10 currencies plus the Korean Won (KRW) from January 1, 2008, to June 30, 2025. The dependent variable is defined as the difference between the synthetic dollar yield on a hedged foreign government bond and the direct dollar yield on a matched-maturity U.S. Treasury.

The proposed benchmark specification is a linear regression of the CIP deviation on three lagged public state variables:

NFCI: The Chicago Fed National Financial Conditions Index (lagged weekly print carried forward to daily frequency).
Broad Dollar: The nominal broad U.S. dollar index (lagged daily observation).
Treasury Slope: The 10-year minus 2-year Treasury yield spread (lagged daily observation).

The paper estimates this model in three configurations:

Separate Currency–Tenor Regressions: Estimating unique coefficients for each currency and maturity.
Currency-Specific Common-Slope: Stacking tenors within each currency, allowing tenor fixed effects but restricting state-variable slopes to be common across tenors.
Pooled Common-Slope: Stacking all currencies and tenors with currency and tenor fixed effects.

Performance is evaluated using within-panel $R^2$ (removing mean differences across panels) and Leave-One-Year-Out (LOYO) out-of-sample tests. The analysis also employs Newey–West HAC standard errors, Engle–Granger cointegration diagnostics, non-overlapping aggregation-difference analysis, and Principal Component Analysis (PCA) to test the robustness and scope of the benchmark.

Key Results

In-Sample Fit: The separate currency–tenor specification yields a within-panel in-sample $R^2$ of 0.637. The currency-specific common-slope specification achieves an in-sample $R^2$ of 0.645, and the pooled common-slope specification reaches 0.658.
Out-of-Sample Performance: The separate currency–tenor specification delivers a LOYO pooled out-of-sample $R^2$ of 0.499. The currency-specific common-slope specification performs even better out-of-sample with a LOYO pooled $R^2$ of 0.607, indicating the fit is not merely an artifact of overfitting unique slopes for every panel.
Coefficient Patterns:
- NFCI: The most stable regressor, with a positive coefficient in 88.8% of panels and significance at the 5% level in 86.7%. This aligns with a limits-to-arbitrage view where tighter financial conditions widen deviations.
- Treasury Slope: Enters positively in 85.7% of panels, suggesting the U.S. yield-curve regime is linked to the persistent component of deviations.
- Broad Dollar: Shows heterogeneous signs, consistent with a global state variable whose impact depends on currency-specific funding and hedging pressures.
Economic Magnitude: A one-standard-deviation increase in NFCI is associated with a median increase of approximately 11.4 basis points in CIP deviations. The Treasury slope has a similar one-standard-deviation effect of roughly 10.0 basis points.
Robustness and Scope:
- Quarter-End Effects: Adding a short-tenor quarter-end dummy (last 5 trading days) adds negligible explanatory power ( $R^2$ change < 0.0001), confirming the benchmark captures background deviations rather than ultra-short maturity reporting spikes.
- VIX Comparison: Replacing NFCI with VIX or adding VIX to the baseline does not improve out-of-sample performance, indicating the benchmark captures broader financial conditions beyond standard equity risk sentiment.
- Cointegration: Engle–Granger tests reject the null of no cointegration in nearly all panels, suggesting the relationship is not a spurious level correlation.
- Frequency: The benchmark explains little of daily high-frequency changes ( $R^2 \approx 0$ ) but becomes increasingly informative at monthly-to-quarterly horizons, confirming it models a persistent macrofinancial component rather than execution noise.
- PCA: The first two principal components of the state variables explain 95% of the variance but fail to recover the benchmark's out-of-sample performance; the full three-variable space is necessary.

Significance and Contribution
The paper does not claim to identify new structural determinants or replace existing mechanism-specific literature. Instead, its primary contribution is benchmarking. It establishes a compact, reproducible, public-data hurdle for future research on daily government-bond CIP deviations.

The paper argues that a small set of three lagged public variables captures a substantial low-frequency component of post-GFC deviations. This specification serves as a "common yardstick" against which richer models (incorporating proprietary balance-sheet data, regulatory variables, or liquidity measures) should be tested. For a new variable or structural model to be considered an improvement, it must demonstrate incremental explanatory power relative to this baseline, both in-sample and out-of-sample, across currencies and tenors. The results suggest that while the benchmark is not a complete structural model, it effectively summarizes the persistent macrofinancial forces driving CIP deviations, providing a necessary reference point for the empirical literature.

The Results: A Surprisingly Good Fit

What This Means for You (The General Audience)

The Bottom Line

Technical Summary: A Three-Variable Benchmark for Post-GFC Covered Interest Parity Deviations

More like this