Wednesday, August 11, 2010

A Good Introduction to GARCH and EWMA (Exponentialy Weighted Moving Average)

GARCH and EWMA


21 May 2010   by David Harper, CFA, FRM, CIPM
AIM:  Compare, contrast and calculate parametric and non-parametric approaches for estimating conditional volatility …
Including: GARCH APPROACH
Including: EXPONENTIAL SMOOTHING (EWMA)

Exponential smoothing (conditional parametric)

Modern methods place more weight on recent information. Both EWMA and GARCH place more weight on recent information. Further, as EWMA is a special case of GARCH, both EWMA and GARCH employ exponential smoothing.

GARCH (p, q) and in particular GARCH (1, 1)

GARCH (p, q) is a general autoregressive conditional heteroskedastic model. Key aspects include:
  • Autoregressive (AR): tomorrow’s variance (or volatility) is a regressed function of today’s variance—it regresses on itself
  • Conditional (C): tomorrow’s variance depends—is conditional on—the most recent variance. An unconditional variance would not depend on today’s variance
  • Heteroskedastic (H): variances are not constant, they flux over time
GARCH regresses on “lagged” or historical terms. The lagged terms are either variance or squared returns. The generic GARCH (p, q) model regresses on (p) squared returns and (q) variances. Therefore, GARCH (1, 1) “lags” or regresses on last period’s squared return (i.e., just 1 return) and last period’s variance (i.e., just 1 variance).
GARCH (1, 1) given by the following equation.
image
The same GARCH (1, 1) formula can be given with Greek parameters: Hull writes the same GARCH equation as:
image
The first term (gVL) is important because VL is the long run average variance. Therefore, (gVL) is a product: it is the weighted long-run average variance. The GARCH (1, 1) model solves for the conditional variance as a function of three variables (previous variance, previous return^2, and long-run variance):
image

Persistence is a feature embedded in the GARCH model.
Tip:  In the above formulas, persistence is = (b + c) or (alpha-1+ beta). Persistence refers to how quickly (or slowly) the variance reverts or “decays” toward its long-run average. High persistence equates to slow decay and slow “regression toward the mean;” low persistence equates to rapid decay and quick “reversion to the mean.”
A persistence of 1.0 implies no mean reversion. A persistence of less than 1.0 implies “reversion to the mean,” where a lower persistence implies greater reversion to the mean.
Tip:  As above, the sum of the weights assigned to the lagged variance and lagged squared return is persistence (b+c = persistence). A high persistence (greater than zero but less than one) implies slow reversion to the mean. But if the weights assigned to the lagged variance and lagged squared return are greater than one, the model is non-stationary. If (b+c) is greater than 1 (if b+c > 1) the model is non-stationary and, according to Hull, unstable. In which case, EWMA is preferred.
Linda Allen says about GARCH (1, 1):
  • GARCH is both “compact” (i.e., relatively simple) and remarkably accurate. GARCH models predominate in scholarly research. Many variations of the GARCH model have been attempted, but few have improved on the original.
  • The drawback of the GARCH model is its nonlinearity [sic]
For example: Solve for long-run variance in GARCH (1,1)
Consider the GARCH (1, 1) equation below:
image
Assume that:
  • the alpha parameter = 0.2,
  • the beta parameter = 0.7, and
Note that omega is 0.2 but don’t mistake omega (0.2) for the long-run variance! Omega is the product of gamma and the long-run variance. So, if alpha + beta = 0.9, then gamma must be 0.1. Given that omega is 0.2, we know that the long-run variance must be 2.0 (0.2 ¸ 0.1 = 2.0).

GARCH(1,1): Mere notation difference between Hull and Allen

image

EWMA

EWMA is a special case of GARCH (1,1) and GARCH(1,1) is a generalized case of EWMA. The salient difference is that GARCH includes the additional term for mean reversion and EWMA lacks a mean reversion. Here is how we get from GARCH (1,1) to EWMA:
image
Then we let a = 0 and (b + c) =1, such that the above equation simplifies to:
image
This is now equivalent to the formula for exponentially weighted moving average (EWMA):
image
In EWMA, the lambda parameter now determines the “decay:” a lambda that is close to one (high lambda) exhibits slow decay.

The RiskMetricsTM Approach

RiskMetrics is a branded form of the exponentially weighted moving average (EWMA) approach:
image
The optimal (theoretical) lambda varies by asset class, but the overall optimal parameter used by RiskMetrics has been 0.94. In practice, RiskMetrics only uses one decay factor for all series:
· 0.94 for daily data
· 0.97 for monthly data (month defined as 25 trading days)
Technically, the daily and monthly models are inconsistent. However, they are both easy to use, they approximate the behavior of actual data quite well, and they are robust to misspecification.
Note: GARCH (1, 1), EWMA and RiskMetrics are each parametric and recursive.

Recursive EWMA

EWMA is (technically) an infinite series but the infinite series elegantly reduces to a recursive form:
image

Advantages and Disadvantages of MA (i.e., STDEV) vs. GARCH

GARCH estimations can provide estimations that are more accurate than MA
image 

Graphical summary of the parametric methods that assign more weight to recent returns (GARCH & EWMA)

image

Summary Tips:

GARCH (1, 1) is generalized RiskMetrics; and, conversely, RiskMetrics is restricted case of GARCH (1,1) where a = 0 and (b + c) =1. GARCH (1, 1) is given by:
image
The three parameters are weights and therefore must sum to one:
image
Tip:  Be careful about the first term in the GARCH (1, 1) equation: omega (ω) = gamma(λ) * (average long-run variance). If you are asked for the variance, you may need to divide out the weight in order to compute the average variance.
Determine when and whether a GARCH or EWMA model should be used in volatility estimation
In practice, variance rates tend to be mean reverting; therefore, the GARCH (1, 1) model is theoretically superior (“more appealing than”) to the EWMA model. Remember, that’s the big difference: GARCH adds the parameter that weights the long-run average and therefore it incorporates mean reversion.
Tip:  GARCH (1, 1) is preferred unless the first parameter is negative (which is implied if alpha + beta > 1). In this case, GARCH (1,1) is unstable and EWMA is preferred.
Explain how the GARCH estimations can provide forecasts that are more accurate.
The moving average computes variance based on a trailing window of observations; e.g., the previous ten days, the previous 100 days.
There are two problems with moving average (MA):
  • Ghosting feature: volatility shocks (sudden increases) are abruptly incorporated into the MA metric and then, when the trailing window passes, they are abruptly dropped from the calculation. Due to this the MA metric will shift in relation to the chosen window length
  • Trend information is not incorporated
GARCH estimates improve on these weaknesses in two ways:
  • More recent observations are assigned greater weights. This overcomes ghosting because a volatility shock will immediately impact the estimate but its influence will fade gradually as time passes
  • A term is added to incorporate reversion to the mean
Explain how persistence is related to the reversion to the mean.
Given the GARCH (1, 1) equation:
image
Persistence is given by:
image
GARCH (1, 1) is unstable if the persistence > 1. A persistence of 1.0 indicates no mean reversion. A low persistence (e.g., 0.6) indicates rapid decay and high reversion to the mean.
Tip:  GARCH (1, 1) has three weights assigned to three factors. Persistence is the sum of the weights assigned to both the lagged variance and lagged squared return. The other weight is assigned to the long-run variance. If P = persistence and G = weight assigned to long-run variance, then P+G = 1.
Therefore, if P (persistence) is high, then G (mean reversion) is low: the persistent series is not strongly mean reverting; it exhibits “slow decay” toward the mean.
If P is low, then G must be high: the impersistent series does strongly mean revert; it exhibits “rapid decay” toward the mean.
The average, unconditional variance in the GARCH (1, 1) model is given by:
image
Explain how EWMA systematically discounts older data, and identify the RiskMetrics® daily and monthly decay factors.
The exponentially weighted moving average (EWMA) is given by:
image
The above formula is a recursive simplification of the “true” EWMA series which is given by:
image
In the EWMA series, each weight assigned to the squared returns is a constant ratio of the preceding weight. Specifically, lambda (l) is the ratio of between neighboring weights. In this way, older data is systematically discounted. The systematic discount can be gradual (slow) or abrupt, depending on lambda. If lambda is high (e.g., 0.99), then the discounting is very gradual. If lambda is low (e.g., 0.7), the discounting is more abrupt.
The RiskMetricsTM decay factors:
  • 0.94 for daily data
  • 0.97 for monthly data (month defined as 25 trading days)
Explain why forecasting correlations can be more important than forecasting volatilities.
When measuring portfolio risk, correlations can be more important than individual instrument volatility/variance. Therefore, in regard to portfolio risk, a correlation forecast can be more important than individual volatility forecasts.
Use GARCH (1, 1) to forecast volatility
The expected future variance rate, in (t) periods forward, is given by:
image
For example, assume that a current volatility estimate (period n) is given by the following GARCH (1, 1) equation:
image
In this example, alpha is the weight (0.1) assigned to the previous squared return (the previous return was 4%), beta is the weight (0.7) assigned to the previous variance (0.0016).
What is the expected future volatility, in ten days (n + 10)?
First, solve for the long-run variance. It is not 0.00008; this term is the product of the variance and its weight. Since the weight must be 0.2 (= 1 - 0.1 -0.7), the long run variance = 0.0004.
image
Second, we need the current variance (period n). That is almost given to us above:
image
Now we can apply the formula to solve for the expected future variance rate:
image
This is the expected variance rate, so the expected volatility is approximately 2.24%. Notice how this works: the current volatility is about 3.69% and the long-run volatility is 2%. The 10-day forward projection “fades” the current rate nearer to the long-run rate.

Compare and contrast the use of historic simulation, multivariate density estimation, and hybrid methods for volatility forecasting.

Nonparametric Volatility Forecasting

image
image

8 comments:

Unknown said...

No author is explaining how weights for alpha , beta and gama are calculated for Garch. In all examples why alpha is .20 where as beta is .7. Why such a big difference. Are weights assigned arbitrarily or there is any method to calculate. Can author throw some light on this.

Unknown said...

In order to calculate the weights alpha, beta and gamma, you need to acquire the returns of the asset the volatility of which you want to model, assume a probability distribution for them, consider their probability density function (normal or log-normal or whatever you think it is) and finally apply Maximum Likelihood Estimation. All you want to do is to maximize the likelihood function (i.e. find the values of the parameters alpha beta and gamma that are most likely to lead to the observations you have already collected). That's the very vague idea, you will find numerous sources online. I hope this helps.

Unknown said...

i would like to know that in GARCH(1,1) Model we are checking ARCH as well as GARCH effect. i will give example of my case here so that you will get my question sir and will help to solve my problem. i have one dependent variable i.e one year bond returns of India and 4 independent variables like interest rate, exchange rate,GDP and deficit finance. in order to calculate mean variance(ARCH effect) we need one dependent variable and one independent variable. can i take dependent variable i.e bond returns and any one independent variable of my choice to run mean variance or i have to follow any rule in order to know which variable out of 4 independent variables shall i use for mean variance and rest 3 independent variables for variance equation(GARCH effect). kindly guide me sir i will be highly obliged to you.

asdfzxh said...

no such thing as good or not, any is ok

Jeff L said...

All of your images (including equations etc) are cropped, regardless of whether the page is viewed in Chrome, Edge, Tor, etc. Which makes reading this article very time consuming. At a guess, it will send many potential eyeballs to other sites.

Jeff L said...

Hull notation and Linda notation are *not* mere differences in notation (I think you got this incorrect information from the Bionic Turtle website). While the ARCH results can sometimes be similar, the methodology is very, very different. One uses squared lagged residuals while the other uses squared lagged returns. That is not a mere notation difference (e.g. I use "alpha" as my squared lagged residual coefficient, while you use "c" as you squared lagged residual coefficient). It is a fundamentally different calculation (e.g. squared lagged residuals are calculated entirely differently to squared lagged returns, and their values will, generally speaking, be very different). Not sure how else to explain the difference, even the formulae you provide on this blog page point to the radically different approaches.

Jeff L said...

The formula for the average, unconditional variance in the GARCH (1, 1) model is missing some brackets. Your graphic uses alpha0 / (1 - alpha1 + beta). It should read alpha0 / (1 - (alpha1 + beta)), or alpha0 / (1 - alpha1 - beta).

Anonymous said...


@Jeff L
1. Right click the pics on a new window.
2. Assuming mean zero, squared returns and residuals coincide. That is the assumption missing, otherwise the notation used is correct overall

What is unclear to me is why does he no Variance forecasts with EWMA.
Clearly, you can do that.