Financial econometrics
The GARCH(1,1) model and volatility dynamics
Financial returns are close to unpredictable in level, but their magnitude is not: turbulent periods follow turbulent periods. Engle's ARCH model1, generalised to GARCH by Bollerslev2, captures this by making today's variance a function of yesterday's shock and yesterday's variance.
The model
Three parameters, one recursion
The shock is a standardised noise scaled by the conditional standard deviation , which is known at date .
| Symbol | Meaning | Unit |
|---|---|---|
| (log) return over period | % per period | |
| conditional mean, constant here | ||
| innovation: unexpected return | ||
| conditional standard deviation (volatility), a function of information at | ||
| i.i.d. noise with zero mean and unit variance |
Today's variance combines a baseline level, yesterday's news (squared shock) and the memory of yesterday's variance. Setting gives back ARCH(1).
| Symbol | Meaning | Unit |
|---|---|---|
| conditional variance of given the past | %² per period | |
| intercept, | ||
| reaction to the previous shock (ARCH effect), | ||
| inertia of the variance (GARCH effect), |
Sources:2
The volatility feedback loop
- Solid arrow: computation at date t
- Dashed arrow: carried over to the next date
- Amber: shock channel (α)
- Blue: memory channel (β)
- Violet: estimation from the data
- level
ω
Sets, together with , the average level of the variance. Must be strictly positive.
- reaction
α
How strongly a recent shock lifts the variance. A high gives a jumpy volatility.
- memory
β
Share of yesterday's variance carried into today. A high gives a smooth volatility.
- persistence
α + β
How fast a variance shock dies out. Close to 1: turbulent episodes last.
Dynamics
Persistence, long-run variance and mean reversion
Take expectations of the recursion using . The formula only makes sense when .
| Symbol | Meaning | Unit |
|---|---|---|
| long-run variance towards which reverts | %² per period | |
| persistence of the variance |
Sources:2
Result derived in this sheet
Obtained by iterating . The gap to the long-run variance decays geometrically; is the number of periods after which it is halved.
| Symbol | Meaning | Unit |
|---|---|---|
| expectation conditional on information available at | ||
| next-period variance, known at | ||
| forecast horizon | periods | |
| half-life of a variance shock | periods |
Forecast variance reverting to its long-run level
View the data
| Series | Forecast horizon h (periods) | Forecast variance / long-run variance |
|---|---|---|
| α + β = 0.90 | 1 | 3 |
| α + β = 0.90 | 3 | 2.62 |
| α + β = 0.90 | 5 | 2.3122 |
| α + β = 0.90 | 7 | 2.0629 |
| α + β = 0.90 | 9 | 1.8609 |
| α + β = 0.90 | 11 | 1.6974 |
| α + β = 0.90 | 13 | 1.5649 |
| α + β = 0.90 | 15 | 1.4575 |
| α + β = 0.90 | 17 | 1.3706 |
| α + β = 0.90 | 19 | 1.3002 |
| α + β = 0.90 | 21 | 1.2432 |
| α + β = 0.90 | 23 | 1.197 |
| α + β = 0.90 | 25 | 1.1595 |
| α + β = 0.90 | 27 | 1.1292 |
| α + β = 0.90 | 29 | 1.1047 |
| α + β = 0.90 | 31 | 1.0848 |
| α + β = 0.90 | 33 | 1.0687 |
| α + β = 0.90 | 35 | 1.0556 |
| α + β = 0.90 | 37 | 1.0451 |
| α + β = 0.90 | 39 | 1.0365 |
| α + β = 0.90 | 41 | 1.0296 |
| α + β = 0.90 | 43 | 1.0239 |
| α + β = 0.90 | 45 | 1.0194 |
| α + β = 0.90 | 47 | 1.0157 |
| α + β = 0.90 | 49 | 1.0127 |
| α + β = 0.90 | 51 | 1.0103 |
| α + β = 0.90 | 53 | 1.0083 |
| α + β = 0.90 | 55 | 1.0068 |
| α + β = 0.90 | 57 | 1.0055 |
| α + β = 0.90 | 59 | 1.0044 |
| α + β = 0.98 | 1 | 3 |
| α + β = 0.98 | 3 | 2.9208 |
| α + β = 0.98 | 5 | 2.8447 |
| α + β = 0.98 | 7 | 2.7717 |
| α + β = 0.98 | 9 | 2.7015 |
| α + β = 0.98 | 11 | 2.6341 |
| α + β = 0.98 | 13 | 2.5694 |
| α + β = 0.98 | 15 | 2.5073 |
| α + β = 0.98 | 17 | 2.4476 |
| α + β = 0.98 | 19 | 2.3903 |
| α + β = 0.98 | 21 | 2.3352 |
| α + β = 0.98 | 23 | 2.2823 |
| α + β = 0.98 | 25 | 2.2316 |
| α + β = 0.98 | 27 | 2.1828 |
| α + β = 0.98 | 29 | 2.136 |
| α + β = 0.98 | 31 | 2.091 |
| α + β = 0.98 | 33 | 2.0478 |
| α + β = 0.98 | 35 | 2.0063 |
| α + β = 0.98 | 37 | 1.9664 |
| α + β = 0.98 | 39 | 1.9282 |
| α + β = 0.98 | 41 | 1.8914 |
| α + β = 0.98 | 43 | 1.8561 |
| α + β = 0.98 | 45 | 1.8222 |
| α + β = 0.98 | 47 | 1.7896 |
| α + β = 0.98 | 49 | 1.7584 |
| α + β = 0.98 | 51 | 1.7283 |
| α + β = 0.98 | 53 | 1.6995 |
| α + β = 0.98 | 55 | 1.6718 |
| α + β = 0.98 | 57 | 1.6452 |
| α + β = 0.98 | 59 | 1.6196 |
Simulate a GARCH(1,1) path
Synthetic data — seed 7with and . The long-run variance exists only when : beyond that, is reduced automatically. Gaussian innovations, zero mean: real returns have fatter tails.
Interactive simulation: a series of simulated returns and, below it, the matching conditional volatility with its long-run level; sliders set omega, alpha, beta and the number of periods, and high-amplitude periods appear grouped together.
- Persistence α + β
- 0.98
- Unconditional variance ω/(1−α−β)
- 2.5
- Unconditional standard deviation
- 1.5811
- Sample variance
- 1.6963
- Shock half-life (periods)
- 34.3
Conditions
Positivity and stationarity
| Property | Condition | Consequence |
|---|---|---|
| Positivity | at every date, whatever the path of shocks. | |
| Covariance stationarity | Finite unconditional variance ; variance forecasts converge2. | |
| Strict stationarity | Weaker than the previous condition: it allows 3. | |
| Finite fourth moment (Gaussian ) | Kurtosis : tails fatter than the normal distribution2. |
Estimation
Quasi-maximum likelihood
The likelihood is maximised as if were Gaussian. If the conditional mean and variance are correctly specified, the estimator remains consistent even when is not; only the standard errors need correcting4.
| Symbol | Meaning | Unit |
|---|---|---|
| parameter vector | ||
| contribution of observation to the log-likelihood | ||
| residual | ||
| variance obtained by running the recursion from a starting value | ||
| number of observations |
Sources:4
Filter the mean
Compute the residuals (or those of a richer mean model).
Initialise and run the recursion
Set , for instance to the sample variance of the residuals, then compute recursively.
Maximise the quasi-likelihood
Numerical optimisation under the constraints , .
Robust standard errors
Use the Bollerslev–Wooldridge "sandwich" covariance matrix, valid without normality4.
Check
Standardised residuals and their squares should no longer be autocorrelated.
Takeaways
Check your understanding
QuestionA GARCH(1,1) has parameters , and . What is its unconditional variance?
- A1
- B0.02
- C0.98
- DIt is infinite
Show the answer
- A — Correct. .
- B — Incorrect. That is , the intercept of the recursion, not the long-run variance.
- C — Incorrect. That is the persistence .
- D — Incorrect. It would be infinite only if .
Confusing conditional and unconditional variance
changes at every date and depends on the past; is a constant. A stationary GARCH is unconditionally homoskedastic and conditionally heteroskedastic.
Believing the model predicts the sign of returns
GARCH describes the expected magnitude, not the direction: whatever .
Reading α + β ≥ 1 as an explosive path
The condition is about the existence of the variance. Strict stationarity obeys a weaker condition3.
Using Gaussian maximum-likelihood standard errors uncorrected
If the standardised innovations are not Gaussian, the inverse Hessian no longer estimates the variance of the estimator correctly: the robust form is required4.
Forgetting the time unit
Parameters depend on the sampling frequency: a GARCH estimated on daily returns does not apply as is to monthly returns.
References
- (1982). Autoregressive Conditional Heteroscedasticity with Estimates of the Variance of United Kingdom Inflation. Econometrica, 50(4), 987. doi:10.2307/1912773
- (1986). Generalized autoregressive conditional heteroskedasticity. Journal of Econometrics, 31(3), 307–327. doi:10.1016/0304-4076(86)90063-1
- (1990). Stationarity and Persistence in the GARCH(1,1) Model. Econometric Theory, 6(3), 318–334. doi:10.1017/S0266466600005296
- (1992). Quasi-maximum likelihood estimation and inference in dynamic models with time-varying covariances. Econometric Reviews, 11(2), 143–172. doi:10.1080/07474939208800229
- (2019). GARCH Models. Wiley. doi:10.1002/9781119313472