This PhD thesis by Wei Chen (University of Birmingham, 2009) explores three distinct but methodologically linked areas of financial and macroeconomic modeling. By applying advanced econometric tools—specifically the GARCH (Generalized Autoregressive Conditional Heteroskedasticity) family of models—the author investigates how risk, volatility, and market pressures behave across different asset classes and geographic regions.
TL;DR
The thesis provides a deep dive into financial risk management, demonstrating that while complex dynamic models (like GARCH) offer superior statistical fit for hedging and volatility transmission, their practical "edge" over simpler models is often context-dependent. Key findings include gold’s unique role as an "exogenous" volatility leader and the effectiveness of China’s oil futures in protecting specifically against downside risk rather than total market variance.
What Problem Does This Paper Address?
The research tackles three primary gaps in the literature of the late 2000s:
- Emerging Market Hedging: At the time of writing, China’s fuel oil futures market was brand new (launched in 2004). There was little empirical evidence on whether these futures actually worked for domestic or cross-border hedging.
- Multivariate Volatility Transmission: Most studies looked at volatility links between just two markets (e.g., oil and stocks). This thesis argues that ignoring a third related market (like gold) creates a "missing variable" bias that obscures the true pathways of financial contagion.
- Crisis Detection Flexibility: Traditional models for predicting currency crises (like Logit/Probit) rely on arbitrary, human-defined "thresholds" to define a crisis. This thesis seeks a more data-driven approach to identify when an economy has switched from a "stable" to a "volatile" regime.
Methods
The common thread across all three essays is the use of GARCH-type models to endogenize market risk.
- Multivariate GARCH (BEKK): Used to model the time-varying covariance between spot and futures prices (Essay 1) and among three different asset classes (Essay 2).
- TGARCH (Threshold GARCH): Employed to capture "asymmetric" effects—the common phenomenon where "bad news" (negative shocks) creates more volatility than "good news."
- Markov Regime-Switching: Used in Essay 3 to allow model parameters to switch between "stable" and "volatile" states automatically, mimicking the existence of multiple equilibria in exchange markets.
- Performance Criteria: Beyond standard variance reduction, the author uses Expected Utility Maximization and Semivariance (downside risk) to evaluate how well these models serve an actual investor.
Datasets
- Essay 1: Daily closing prices for China fuel oil futures (SHFE), domestic spot prices (Huangpu), and Singapore spot prices (Platts 180CST) from 2004 to 2006.
- Essay 2: Daily returns for NYMEX Light Sweet Crude Oil, CME Gold futures, and the S&P 500 index (1991–2007).
- Essay 3: Monthly macroeconomic data (reserves, real exchange rates, domestic credit) for Korea, Indonesia, Thailand, Malaysia, Singapore, and the Philippines (1980–2008).
Key Insights and Ideas
- The "Third Market" Matters: In Essay 2, the author finds that relationships that seem significant in a bi-variate (two-market) model often disappear or change when a third market is added. For example, the apparent link between oil and gold volatility is actually driven by their mutual relationship with the stock market.
- Gold as the Volatility Leader: Gold is found to be "exogenous" in terms of its second moment. Its volatility affects both the oil and stock markets, but it is largely unaffected by them, reinforcing its status as a "safe haven" that dictates terms to other assets during turmoil.
- Downside Protection vs. Total Hedging: A major "aha moment" in Essay 1 is that China’s oil futures are actually quite poor at reducing total variance (total risk) but are highly effective at reducing downside risk (semivariance). For an investor who only fears losses but wants to keep upside potential, the futures are more useful than they appear on paper.
- Regime-Switching Superiority: In Essay 3, Markov Switching models proved more flexible than Logit models because they could capture "appreciating" currency attacks (where a currency is forced upward) as well as the more common "depreciating" crashes.
Key Results
- Hedging Effectiveness: In-sample, dynamic GARCH models reduced variance by about 20-25%. However, out-of-sample performance dropped significantly to 3-9%, suggesting that GARCH models are highly sensitive to the specific time period and struggle with long-term forecasting.
- Crisis Prediction: The Markov Switching TGARCH model correctly identified the 1997 Asian financial crisis in all six countries, often detecting the shift to a "volatile" state a month before the actual crash.
- Model Comparison: While the complex TGARCH models were statistically superior (based on Likelihood Ratio tests), the simpler Multinomial Logit models actually provided higher "correct prediction" percentages for depreciating attacks, though they were less theoretically robust.
Limitations and Future Work
- Data Constraints: Essay 1 was limited by a very short history of the Chinese oil futures market (only two years).
- Forecasting Decay: The author acknowledges that GARCH models require frequent updates; without constant re-estimation, their out-of-sample predictive power for hedging ratios decays rapidly.
- The "Third Generation" Gap: Essay 3 focuses on macroeconomic fundamentals (1st and 2nd generation crisis models) but suggests that future work should incorporate "3rd generation" factors like banking sector fragility and moral hazard (e.g., government bailouts).
- Three-Regime Models: The author suggests that future currency research should move from two regimes (stable/volatile) to three (stable/appreciating/depreciating) to better capture market nuances.
Summary updated