📈 economics

The Regime Switch of Capital Flows: How the China-U.S. Yield Spread Drives Global Capital Reallocation in a Tiered Manner— Exhaustive Evidence Based on TIC Panel Data, ETF Flows, and China's Cross-Border Data

This paper proposes and empirically validates a "Regime Switch Theory" of capital flows, demonstrating that the China-U.S. yield spread acts as a structural threshold that determines whether global capital allocation is driven by U.S. market factors or gains independent pricing power through China-specific variables, thereby unifying diverse capital flow phenomena into a single "Unified Field" framework.

Original authors: Shuiping Tang

Published 2026-07-14

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Shuiping Tang

Original paper licensed under CC BY 4.0 (https://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 economy as a massive, chaotic game of musical chairs where trillions of dollars are constantly running around looking for the best seat. For forty years, economists have been trying to figure out what music is playing. Their old theory was simple: the music is just a steady beat. If the interest rate in one country goes up by a tiny bit, money just flows there in a straight, predictable line, like water filling a cup.

But this paper, written by independent researcher Shuiping Tang, says that old theory is missing the most important part of the game: the DJ.

The DJ and the Dance Floor

Tang argues that the "China-U.S. yield spread" (the difference between interest rates in China and the U.S.) isn't just another dancer on the floor. It's the DJ who decides which genre of music is playing.

When the spread is positive (U.S. rates are higher), the DJ switches the music to "American Style." Suddenly, the most important factors driving money are American things like the VIX (a panic meter) and the Dollar Index.
When the spread is negative (China's rates are higher), the DJ flips the switch to "Chinese Style." Now, Chinese factors, like the RMB exchange rate, take the lead and start directing the flow of cash.

The paper's main finding is that the spread doesn't push the money directly. Instead, it modulates the volume of other factors. Think of it like a volume knob. The spread doesn't make the music; it just turns the volume up on the American band or the Chinese band depending on the setting.

The "Rule-Maker" vs. The "Rule-Follower"

The paper explicitly rules out the idea that all economic factors are equal. It says we've been confusing "rule-makers" with "rule-followers."

Rule-followers are things like stock market panic or currency values. They just do what they're told.
Rule-makers are the interest rate spreads. They decide which rule-followers get to speak.

The authors tested this by looking at data from 1978 to 2023, covering 68 countries, stock markets, bond markets, and even ETF trading volumes. They used twelve different statistical methods to try and break their own theory, but the pattern held up every time.

The "Tiered" Reaction: Who Moves When?

Here is where it gets really interesting. The paper found that different types of money move to the music in completely different ways.

Official Money (The Strategists): This is money from central banks and governments. When the spread turns negative (U.S. rates drop relative to China), these big players stop buying U.S. Treasuries. They aren't just reacting to profit; they are playing a long-term game of "de-dollarization."
- The Proof: Among the top 10 countries holding U.S. debt, China is the only one that actually reduces its holdings when the spread widens. Its coefficient is -153.55, meaning it's doing the opposite of everyone else who is just chasing profit.
Private Money (The Tacticians): This is money from regular investors and funds. They are more like gamblers. When the spread inverts, they get scared and run back to the "safe" dollar, but they don't blindly follow the same rules as the central banks.

The paper also found a "speed limit" for different assets.

Stocks (Equities): These are fast cars. They react almost instantly when the DJ changes the music. The statistical test (Chow test) for stock flows showed a massive difference between positive and negative spreads, with an F-value of 33.15 (p=0.000).
Bonds: These are slow-moving trucks. The paper found that for foreign bond holdings, the "DJ switch" barely matters. The statistical test was insignificant (F=0.22, p=0.928). This means bond investors are so focused on the long term that they don't care about the monthly music changes.

The "Institutional Interface" (The Doorbell)

The paper also explains why this works for some assets and not others. It's all about the "Institutional Interface."

The Doorbell: Imagine a house. If you don't have a doorbell, people can't ring to get in. In finance, you need a specific "door" (like the Shanghai-Hong Kong Stock Connect) for foreign money to easily enter a market.
The Proof: When the Stock Connect was launched in 2014, the "DJ" (the spread) suddenly started controlling the flow of money into Chinese stocks. Before the doorbell was installed, the spread was silent.
The Counter-Proof: The paper looked at the MCHI/SPY volume ratio (a comparison of a China ETF and a U.S. ETF). The spread had zero effect on this ratio (p=0.875). Why? Because the China ETF trades in the U.S. without a Chinese "doorbell." Without the interface, the signal can't get through.

What About the Old Theories?

The paper doesn't say the old theories (like the Mundell-Fleming model or the Triffin Dilemma) are "wrong." Instead, it says they are special cases that only work under very specific, limited conditions.

It's like saying Newton's laws of physics are great for everyday objects, but they break down when you get to the speed of light.
The old theories assume the "DJ" never changes the music. This paper proves the DJ changes the playlist constantly, and once you account for that, the old models fall apart.

How Sure Are They?

The authors are very confident, but they are careful. They didn't just guess; they ran exhaustive tests.

They tested 462 and 141 factor combinations in other markets (gold and oil) to find this pattern.
They used 200 Bootstrap iterations (a fancy way of simulating the data thousands of times) to prove the "switch" happens right around a spread of 0.
They found that the "interaction effect" (the DJ turning the volume knob) is 1.78 times stronger than the direct effect.
They even checked if the results were just a fluke by shuffling the data randomly 200 times, and the real pattern was still way stronger (p=0.000).

The Bottom Line

This paper suggests that for forty years, economists have been trying to predict the weather by looking at the temperature, missing the fact that the season (the regime) determines what the temperature actually means.

When the China-U.S. spread is positive, the U.S. sets the rules. When it's negative, China sets the rules. And the most surprising part? The "official" money (governments) and "private" money (investors) are dancing to completely different beats, even when the music is the same. The paper completes a massive puzzle, connecting everything from gold prices to stock markets, proving that the global economy isn't a single machine, but a dual-anchor system that flips its logic based on a simple switch.

Technical Summary: The Regime Switch of Capital Flows

1. Problem Statement

The international capital flow literature has long operated under the presumption that macroeconomic drivers, such as interest rate differentials, function as linear, stable factors with constant exposure coefficients. This paper challenges that foundational assumption, arguing that the China-U.S. yield spread is not a standard pricing factor but a regime variable.

The core problem identified is a methodological blind spot: existing literature treats "rule-makers" (regime factors that determine which logic dominates) and "rule-executors" (market factors like VIX or DXY) as parallel variables. The paper posits that the China-U.S. yield spread determines which side's logic (U.S.-side or China-side) currently drives global capital reallocation. When the spread is positive, U.S. factors dominate; when negative, China-side factors gain independent pricing power. Traditional linear models fail because they average out these opposing regime effects, rendering macro factors statistically insignificant.

2. Theoretical Framework: The Unified Field Formula

The paper derives a Unified Field Formula for global capital flows based on the institutional fact that central banks independently set interest rates under a fiat currency system:

$F_t = \Sigma w_i(S_t) \times f_i(X_t^i) + \varepsilon_t$

Where:

$F_t$ : Capital flows (direction and magnitude).
$S_t$ : The regime variable (currently the China-U.S. yield spread).
$w_i(S_t)$ : A weight function determined by the regime variable, dictating the influence of each economy's pricing logic.
$f_i(X_t^i)$ : The pricing function of the $i$ -th economy, driven by market factors (e.g., VIX, DXY).
Institutional Interface: The condition allowing a country's macro variables to enter the global equation (e.g., Stock Connect, Bond Connect).

Core Propositions:

Interaction Effect: Regime factors do not directly price flows (direct effect $\approx 0$ ) but indirectly price them by modulating the exposure coefficients of market factors via interaction terms.
Threshold Effect: A structural break occurs near a spread of zero ( $S_t \approx 0$ ), where the pricing logic switches.
Tiered Driving: Official capital (strategic de-dollarization) and private capital (tactical risk aversion) respond with opposite logic to the spread.
Institutional Interface Activation: The pricing function of the spread is "silent" without an institutional interface and "activated" once one is established.

3. Methodology

The paper employs an exhaustive testing approach designed to systematically exclude alternative hypotheses rather than merely confirm a specific model. The analysis covers the period from January 2005 to May 2023 (221 monthly observations) using the following data sources:

TIC Data: 68-country panel (1978–2023) for official and private holdings of U.S. Treasuries, equities, and bonds.
China Flows: Northbound/Southbound flows (Stock Connect) and foreign holdings of four Chinese asset categories (A-shares, bonds, loans, deposits).
ETF Flows: Volume data for eight major country ETFs (including MCHI and SPY).
Macro Factors: VIX, DXY, exchange rates, and yield spreads (10Y, 30Y-10Y).

Twelve-Layered Testing Framework:

Exhaustive Factor Combinations: Traversing all single, two, and three-factor combinations under a Variance Inflation Factor (VIF) < 5 constraint to find optimal pricing models.
Interaction Effect Testing: Regressing flows against regime factors, market factors, and their interaction terms.
Regime Grouping: Estimating coefficients separately for positive and negative spread regimes.
Chow Tests: Testing for structural breaks in coefficients between regimes.
Threshold Regression: Using Bootstrap (200 iterations) to identify significant thresholds near $S_t \approx 0$ .
Rolling Window Regression: Observing coefficient evolution over 3-year and 5-year windows to detect regime switches.
Tiered Driving Analysis: Comparing official vs. private capital responses.
Institutional Event DID: Using the 2014 launch of the Shanghai-Hong Kong Stock Connect as a natural experiment.
Fama-MacBeth Two-Stage Regression: Analyzing the cross-sectional distribution of spread coefficients across 68 countries.
COFER Lag Tests: Examining long-term (4–8 quarter) impacts on reserve currency shares.
Quantile Regression: Testing for asymmetry in extreme quantiles.
Robustness Checks: Including LASSO regularization, permutation tests, Newey-West corrections, and alternative hypothesis exclusion (e.g., adding Japanese/German spreads).

4. Key Results

A. The Interaction Effect (Prediction 1)

Direct vs. Interaction: The direct effect of the spread on capital flows is statistically insignificant (e.g., $p=0.969$ for official U.S. Treasuries after Newey-West correction). However, the interaction term (Spread $\times$ Market Factor) is highly significant.
Magnitude: The incremental $R^2$ from interaction terms is 1.78 times that of the direct effect.
Universality: This pattern holds across aggregate flows, six categories of disaggregated market flows (Treasuries, Equities, Corporate/Agency Bonds), and six major equity market volatility measures.

B. The Threshold Effect (Prediction 2)

Structural Breaks: Chow tests confirm significant structural differences between positive and negative spread regimes for nine out of ten capital flow categories.
Threshold Location: Threshold regressions identify the critical switching point near spread $\approx 0$ (specifically 0.008–0.018 for core variables), consistent with findings in commodity markets.
Asset Specificity: The bond market (foreign holdings of Chinese bonds) is an exception, showing an insignificant Chow test ( $F=0.22$ ). This establishes a theoretical boundary: bond flows (long-term allocation) react far slower than equity flows (fast variables).

C. Tiered Driving Mechanism (Prediction 3)

Official vs. Private: Official capital exhibits "strategic de-dollarization," ceasing to add to U.S. Treasuries when the spread turns negative. Private capital exhibits "tactical risk aversion," often reversing signs or persistently reducing U.S. equity holdings.
Country Heterogeneity: In the 68-country TIC panel, China is the only top-10 holder with a negative spread coefficient (-153.55). While most countries increase U.S. Treasury holdings when the spread widens (profit-driven), China reduces them, indicating strategic reserve diversification.

D. Institutional Interface Verification (Prediction 4)

Activation: The launch of the Shanghai-Hong Kong Stock Connect significantly enhanced the marginal effect of the spread on capital allocation (DID results significant at $p < 0.01$ ).
Absence Verification: The trading volume ratio of MCHI (China ETF in the U.S.) to SPY is not driven by the spread ( $p=0.875$ ). This confirms that without a local-currency institutional interface, the spread cannot price the asset, validating the "Institutional Interface" theory.

E. Replication Across Dimensions

The paper replicates the "direct effect $\approx 0$ , interaction effect significant" pattern found in previous work on prices (Tang, 2026e) within the quantity dimension of capital flows. This cross-dimensional consistency is presented as an inevitable deduction from the Unified Field Formula.

5. Significance and Claims

The paper claims to complete the "Dual-Anchor Regime" research program, moving from price discovery to quantity analysis. Its primary contributions are:

Theoretical Reduction: It does not claim to overthrow existing theories (Mundell-Fleming, Triffin Dilemma, traditional capital flow models) but reduces them to special cases of the Unified Field Formula under specific boundary conditions (e.g., when institutional interfaces are absent or the regime variable is constant).
Methodological Correction: It argues that the marginalization of macro factors in 40 years of literature is due to the use of static linear regressions, which average out opposing regime effects. The paper demonstrates that regime switches render these tools inappropriate.
Dual-Anchor Reality: It provides empirical evidence that the global pricing system has shifted from a single U.S. dollar anchor to a dual China-U.S. anchor, where the relative stance of monetary policies (the spread) dictates which anchor dominates.
Closed Loop: Together with five preceding papers (Tang, 2026a–2026e), this work forms a complete research program covering commodities, fixed income, currencies, equities, and capital flows, establishing a "Unified Field of Spreads" that governs both price and quantity dimensions.

The paper concludes that the core insight is not the discovery of a new factor, but the recognition that pricing factors are hierarchical: some set the rules (regime factors), while others execute within those rules (market factors). Confusing these two roles is identified as the primary methodological error in the field.