If you’ve ever watched the stock market, you know it can feel like a roller coaster. But it’s not just any roller coaster; it’s a moody one. It seems to overreact to bad news-a poor earnings report, a regulatory scare, or a global crisis-with sharp, stomach-churning drops. Yet, when good news hits, the climb up often feels more gradual, almost cautious. This isn’t just a feeling; it’s a well-documented phenomenon in financial markets. We’re talking about volatility, the *speed and size* of price changes, and the fact that it’s not symmetrical.
For decades, economists and financial analysts have tried to model this “storminess.” One of the most famous tools in their kit is the GARCH model. GARCH, which stands for Generalized Autoregressive Conditional Heteroskedasticity (a mouthful, we know), is brilliant at one key thing: modeling volatility clustering. This is the simple observation that calm days are often followed by calm days, and chaotic, volatile days are followed by more chaotic days. GARCH looks at yesterday’s volatility and yesterday’s “news” (the market shock) to predict today’s volatility.
But the standard GARCH model has a significant blind spot. It’s “colorblind” to the *type* of news. It treats the volatility impact from a 2% market gain (good news) as exactly the same as the volatility impact from a 2% market drop (bad news). And as we’ve already noted, that’s just not how the real world works. This is where we run into a critical concept: the leverage effect.
Table of Contents
What exactly is the leverage effect?
The leverage effect is the name for this asymmetric relationship: negative shocks (bad news) tend to increase future volatility far more than positive shocks (good news) of the same magnitude. This idea was first explored in-depth by economist Fischer Black in 1976, and it’s a cornerstone of modern financial modeling.
But why does this happen? There are two main theories that are likely working together:
- The Financial Leverage Story: This is where the name comes from. Imagine a company has a certain amount of debt and a certain amount of equity (its stock value). This mix is its “leverage.” When the company’s stock price falls (bad news), its equity value shrinks, but its debt stays the same. This automatically increases its debt-to-equity ratio, making the company financially more leveraged. A more leveraged company is inherently riskier, and this increased risk perception leads directly to higher volatility in its stock price.
- The Volatility Feedback Story: This theory is more about human psychology. Bad news, like a poor economic forecast or a political scandal, creates genuine fear and uncertainty. Fear often leads to panic selling, which itself increases volatility. This increased volatility makes *other* investors nervous, leading to more selling. It’s a feedback loop. Good news, on the other hand, is often met with calmer, more orderly profit-taking, not a “panic-buying” spree.
Whatever the cause, this effect is real. Studies on major indices, including those in India like the Nifty 50, consistently show that negative returns are followed by bigger spikes in volatility than positive returns. The standard GARCH model can’t capture this, which means its risk forecasts will be wrong. It will underestimate risk on days after bad news and potentially overestimate it after good news. To fix this, economists developed a whole new “family” of GARCH models.
Meet the specialists: Three advanced GARCH models
If the standard GARCH is a general practitioner, think of EGARCH, TGARCH, and PGARCH as the specialists called in to handle this complex asymmetric problem. They are all designed to do one thing: build the leverage effect directly into the model.
1. The Exponential GARCH (EGARCH) model
Proposed by Daniel Nelson in 1991, the Exponential GARCH (EGARCH) model was a revolutionary approach. Its design cleverly solves two problems at once.
The Big Idea: Model the log, not the level
Instead of modeling the conditional variance ($\sigma_t^2$) directly, EGARCH models its natural logarithm ($log(\sigma_t^2)$). This might seem like a small mathematical trick, but it has a huge practical advantage. Since the logarithm of a number can be positive, negative, or zero, the model’s parameters don’t need any restrictions to be estimated. And because the *log* of the variance can be negative, the variance *itself* (which is the exponent of the log) will always be positive. This neatly solves a major headache of standard GARCH, which required parameter constraints to prevent the model from ever predicting a negative (and thus impossible) variance.
The Big Payoff: Capturing asymmetry
The real magic is in how the EGARCH equation is structured. It includes a term that explicitly separates the *size* of the news from the *sign* of the news.
Think of it this way:
- One part of the equation looks at the magnitude of the shock: “How big was the surprise?” (e.g., 2%).
- A second, crucial part looks at the sign of the shock: “Was that 2% positive or negative?”
The model includes a leverage coefficient (gamma, $\gamma$). If this coefficient is found to be negative and statistically significant, it confirms the leverage effect. It means that a negative shock (bad news) has a larger impact on the log-variance than a positive shock (good news) of the same size. The “exponential” nature of the model means this asymmetric effect is, well, exponential, allowing for a very strong response to bad news.
2. The Threshold GARCH (TGARCH) model
If EGARCH is the mathematically elegant solution, the Threshold GARCH (TGARCH) model is the brilliantly simple and direct one. It’s also known as the GJR-GARCH model, named after its creators Glosten, Jagannathan, and Runkle (1993).
The Big Idea: An “on/off” switch for bad news
The TGARCH model works by adding a “threshold” term to the standard GARCH equation. This new term is connected to a “dummy variable,” which acts like a light switch.
- This switch is “off” (the dummy variable is 0) if the past news was good (a positive market shock).
- The switch flips “on” (the dummy variable is 1) if the past news was bad (a negative market shock).
This “on” switch activates an additional coefficient (gamma, $\gamma$).
The Big Payoff: A clear, additive effect
The beauty of this is its simple interpretation.
- When news is good: The switch is off. The impact of the shock on volatility is just the standard GARCH amount (let’s call it $\alpha$).
- When news is bad: The switch is on. The total impact of the shock on volatility is the standard amount *plus* the extra “bad news” amount ($\alpha + \gamma$).
If that $\gamma$ coefficient is positive and significant, it is direct, unmistakable evidence of the leverage effect. It’s literally measuring the *additional* volatility that bad news creates over and above good news. This intuitive “bad news dummy” approach has made TGARCH an extremely popular choice for financial analysts. Many studies modeling volatility in emerging markets, including India, find that the TGARCH model provides a superior fit to the standard GARCH model, proving how pervasive the leverage effect is.
3. The Power GARCH (PGARCH) model
Finally, we have the Power GARCH (PGARCH) model, developed by Ding, Granger, and Engle (1993). If the other models are specialists, PGARCH is the “Swiss Army knife”-it’s designed for maximum flexibility.
The Big Idea: Let the data decide the power
The PGARCH model generalizes the standard model in a very clever way. Remember how standard GARCH models the *variance* (which is the standard deviation squared, $\sigma_t^2$)? The PGARCH model introduces a new parameter, a power `d` (delta), and models $\sigma_t^d$.
Why is this cool? Because the model *estimates* the best value for `d` from the data.
- If the data suggests `d=2`, the model just becomes a standard GARCH model (with a leverage term).
- If the data suggests `d=1`, the model is actually modeling the *standard deviation* ($\sigma_t$) directly, which some research suggests is a better, more stable measure of volatility.
- The data could even decide that the best fit is `d=1.5` or some other value.
This flexibility allows the PGARCH model to capture different facets of volatility behavior that other models might miss.
The Big Payoff: Flexibility plus leverage
Of course, the PGARCH model *also* includes a leverage term (a $\gamma$ coefficient) just like the other models, allowing it to capture the asymmetric impact of good and bad news. This combination makes it an incredibly robust and general framework. It doesn’t force the researcher to assume that modeling variance (power of 2) is the right way to go. It allows for a more nuanced investigation, asking both “Is there a leverage effect?” and “What is the true underlying power of volatility that we should be modeling?”
Which model is best, and why does this matter?
So, we have three powerful tools: EGARCH (the log-based model), TGARCH (the threshold model), and PGARCH (the flexible power model). Which one is the “best”?
The answer is: it depends entirely on the data. There is no single champion. An analyst at the Reserve Bank of India (RBI) modeling bond yield volatility might find that an EGARCH model provides the best forecasts. A hedge fund quant modeling the Nifty 50 might find that the simple, intuitive TGARCH model fits the data perfectly. A researcher studying cryptocurrency volatility might prefer the robust, flexible PGARCH model to handle extreme price movements.
The job of the economist is to test these models against each other and see which one best captures the “personality” of the specific asset they are studying. But the critical takeaway is that simply ignoring the leverage effect is no longer an option. The standard GARCH model, while foundational, is built on an assumption of symmetry that is clearly violated in the real world.
This isn’t just an academic debate. For a risk manager, underestimating volatility after a market crash (as a standard GARCH model would) could be catastrophic. For an options trader, accurately pricing an option depends almost entirely on forecasting future volatility; getting the asymmetry right is the difference between profit and loss. These advanced models give us a clearer, more accurate lens to understand financial risk, one that finally acknowledges the simple truth: in markets, bad news hits a whole lot harder than good news.
What do you think? Have you ever noticed this “leverage effect” in your own experience, where bad news seems to cause a much stronger, more sudden reaction in the market than good news? If you were managing a portfolio, why would understanding this asymmetry be more important than just knowing the *average* level of volatility?
Leave a Reply