π ππ‘ = ββππ=1ln (βπβ ππ)2
4ππ(2)
where π ππ‘ is the indexβ realised volatility, βππ=1ln (βπ β ππ)2is the sum of natural loga-rithm of the difference between highest and lowest price during the sample period and T is number of days in the sample period, which in this research is two for daily volatility and 22 for monthly volatility.
Following Christensen and Prabhala (1998), the option implied volatility is calculated us-ing Black and Scholes (1973) option pricus-ing formula for European call options. The option implied volatility is calculated daily for the MSCI Emerging Market Price Index Options as the option prices, underlying index price and risk-free rate are known. As presented in chapter 3.1.1, the Black-Scholes model the option implied volatility is calculated from call option formula as follows:
π0 = π0π(π1) β πΎπβπππ(π2)
where
π1 = ln (π0βπΎ)+ (π + π2β2)π
πβπ
π2 = ln (π0βπΎ)+ (π β π2β2)π
πβπ = π1β πβπ
where π0 is current call option value, π0 is current stock price, N(d1) is the factor by which the present value of a random price of the stock exceeds the current stock price, N(d2) is the probability of the option being exercised, πΎ is the option exercise price, e (Napierβs constant) is a constant and the base of the natural logarithm function, r is the risk-free interest rate which is the 3-month T-bill rate, T is the time to optionβs expiration in years, ln is the natural logarithm function. The implied volatility is represented by Ο,
which is the standard deviation of the annualized and continuously compounded rate of return on the underlying stock:
πΌππ‘= ππ‘ (24)
where πΌππ‘ is the option implied volatility measure for day t and ππ‘is the calculated value.
The daily implied volatility is computed for the entire data period. The value calculated from the Black-Scholes option pricing formula provides an annualised volatility. In order to calculate the daily volatility, the value is divided with β252 as follows:
πΌππππππ¦ = πt / β252 (25)
πΌπππππ‘βππ¦ = πΌππππππ¦β β22 (26)
The implied volatility measure is computed using a VBA macro (Learn365 Club, 2019) that loops Excelβs goal seek function for each row:
Sub Goal_Seek_Range_MultipleGoal() 'Defining variable k
Dim k As Integer
'Looping through each row of the table For k = 2 To 1829
'Replicate the Goal Seek function via VBA
Cells(k, "M").GoalSeek Goal:=Cells(k, "B"), Chang-ingCell:=Cells(k, "F")
'Go to next iteration Next k
End Sub
Following Gokcan (2000), the GARCH(1,1) model is computed as follows:
ππ‘2 = π + β πΌππ’π‘βπ2
where ππ‘2 is the GARCH modelled variance (GV) at time t, π is the long-run average var-iance rate, π’π‘βπ2 is the lagged time point t-i return and πΌπ is the weight assigned to each value in the sum, ππ‘βπ2 is the lagged time point t-j variance term and π½π is the weight as-signed to each value in the sum. The Ο, Ξ± and Ξ² are positive constant parameters and π + Ξ± + Ξ² = 1.
The daily and monthly GARCH(1,1) volatilities are then calculated with π‘ β π and π‘ β π being one-day lag and monthly forecast is computed by multiplying the daily volatility with β22.
Following the methods of Hull (2011, pp. 528), the coefficients Ο, Ξ± and Ξ² are defined for the whole sample period using maximum likelihood estimation method. By maximis-ing the likelihood of data occurrmaximis-ing, the followmaximis-ing equation is maximised usmaximis-ing Excel number of observations occurring. The Excel Solver is used to maximise the GARCH(1,1) coefficients with restrictions that 0 β₯ Ο β₯ 1, 0 β₯ Ξ± β₯ 1 and 0 β₯ Ξ² β₯ 1. The maximum likeli-hood estimation method provides the coefficient values in Table 2 that are used to cal-culate the GARCH(1,1) volatility values daily for the entire sample period:
Table 2. GARCH(1,1) coefficient values.
5.2.2 OLS Regressions
Following Dutta (2017), the OLS regressions to examine the predictive power of implied volatility (IV) and GARCH(1,1) modelled volatility (GV) for daily and monthly volatility forecasts are computed as follows:
π ππ‘+1= πΌ0+ π½1πΌππ‘+ ππ‘+1 (30)
π ππ‘+1= πΌ0+ π½1πΊππ‘+ ππ‘+1 (31)
where RVt+1 indicates the 1-day ahead realised volatility. The OLS models are tested for the entire time period of 1.1.2015β31.12.2019.
By testing the null hypothesis, whether H0 : Ξ²1 = 0, the results indicate if the volatility models include information on future realised volatility. If the coefficient Ξ²1 is statistically different from zero, the results show evidence that implied volatility and GARCH(1,1) have significant predictive power over future emerging stock market volatility. (Dutta, 2017)
Coefficient Value
Ο 0.000000998
Ξ± 0.083
Ξ² 0.91
Note: The values of Ο, Ξ± and Ξ² represent the values of maximum likelihood estimation of GARCH(1,1) parameters.
5.2.3 Error Terms (RMSE & MAE)
The examination of forecasting accuracy includes also the examination of residuals that are the difference between the forecasted value and the observed value. Following Dutta (2017), two error statistics, Root Mean Square Error (RMSE) and Mean Average Error (MAE), are computed to determine the error between the regression line and realised volatility.
The Root Mean Square error (RMSE) is the standard deviation of the residuals. It measures how far the residuals are from the regression lineβs data points when regres-sion line is the best fit among the forecasted values. RMSE indicates the mean of squared differences between the actual volatility values and forecasted volatility values as follows:
π πππΈ = β1
πβππ=1(π ππ,π‘ β π ππ,π‘)2 (32)
where βππ=1(π ππ,π‘β π ππ,π‘)2 is the sum of squarer between the forecasted values π ππ,π‘ and actual values π ππ,π‘ and N is the number of observations. (Barnston, 1992;
Dutta, 2017)
Another error measure is the Mean Absolute Error (MAE), which measures the differ-ence between two continuous variables. In this study, following Dutta (2017), MAE is used to measure the error between the observed volatility value and forecasted value.
MAE is calculated for each predicted and realised value. In this case the error of the forecast is computed the following way:
ππ΄πΈ = 1
πβ |π ππ,π‘ β π ππ,π‘|
π π=1
(33)
where βππ=1|π ππ,π‘β π ππ,π‘| is the sum of absolute difference between the actual value π ππ,π‘ and the forecasted value π ππ,π‘ and N is the number of observations. (Dutta, 2017;
Wilmott & Matsuura, 2005)