Heteroskedastic Models Submitted By: Soumya Sreekumar 12DM-145 Arifa Kazi 12FN-024 Prakhar Saxena 12DCP-080 Nikash Kumar 12DCP-071 Tanuj Arora 12DCP-117 Deepanshu Siddhanti 12DCP-029 Agenda Data Historical Exchange Rates for 3 years (September 2010 -2013)
Tests for Stationarity If non stationary data exists, conversion to stationary
Applying different Models ARCH, GARCH, TARCH, E-GARCH and PARCH
Initial Approach Exchange Rates Time Series Data It can have Variable Variance and Far off Mean indicating Non Stationarity Thus, Cannot be Modeled or Forecasted as it is Unit Root Testing in Eviews determines Non Stationarity Conversion to Stationarity by Taking Log Normal Returns
Unit Root Tests Phillips Peron Test
Augmented Dickey-Fuller Test
P-value > 0.05 means Null Hypothesis cannot be rejected.
Conclusion: Exchange Rate data is non - stationary Conversion to Stationary Data Eviews Quick Tab Generate Series Equation entered is: usd_s = dlog(inr_usd) to give Continuously Compounded Returns. Applying Dickey Fuller Test
P-value < 0.05 means Null Hypothesis can be rejected
Conclusion: Exchange Rate Log Return Data is stationary ARMA Estimation Output Eviews Quick Tab Estimate Equation Equation: Dlog(inr_usd) c ar(1) ma(1)
P-value < 0.05 means Null Hypothesis can be rejected
Conclusion: Model is significant AR Yesterdays Price does not have a negative impact on current price MA Yesterdays Volatility has a positive impact on current price ARMA Graph
Garch Aim Forecast variance based on past information
All coefficients are positive and significant
Shocks to volatility have a persistent effect on the conditional variance
Tarch Aim To check asymmetric news impact on the volatility of exchange rate
Gamma is negative and significant
Bad news may not increase volatility Gamma 0 implies asymmetric effect of news is present
Dependent Variable: DLOG(INR_USD) Method: ML - ARCH (Marquardt) - Normal distribution Date: 09/11/13 Time: 20:48 Sample (adjusted): 9/14/2010 9/10/2013 Included observations: 763 after adjustments Convergence achieved after 15 iterations MA Backcast: 9/13/2010 Presample variance: backcast (parameter = 0.7) GARCH = C(4) + C(5)*RESID(-1)^2 + C(6)*RESID(-1)^2*(RESID(-1)<0) + C(7)*GARCH(-1) Variable Coefficient Std. Error z-Statistic Prob. C 0.000264 0.000192 1.375826 0.1689 AR(1) -0.555827 0.168779 -3.293217 0.0010 MA(1) 0.671094 0.152369 4.404401 0.0000 Variance Equation C 7.90E-07 2.48E-07 3.189087 0.0014 RESID(-1)^2 0.145466 0.030221 4.813429 0.0000 RESID(-1)^2*(RESID(-1)<0) -0.098825 0.026631 -3.710951 0.0002 GARCH(-1) 0.883509 0.022334 39.55864 0.0000 R-squared 0.014542 Mean dependent var 0.000423 Adjusted R-squared 0.006720 S.D. dependent var 0.005930 S.E. of regression 0.005910 Akaike info criterion -7.670058 Sum squared resid 0.026407 Schwarz criterion -7.627514 Log likelihood 2933.127 Hannan-Quinn criter. -7.653678 F-statistic 1.859278 Durbin-Watson stat 2.008150 Prob(F-statistic) 0.085233 Inverted AR Roots -.56 Inverted MA Roots -.67 E-Garch Variable Coefficient Std. Error z-Statistic Prob. C 0.000177 0.000175 1.010816 0.3121 AR(1) -0.596517 0.143221 -4.165017 0.0000 MA(1) 0.711246 0.124667 5.705174 0.0000 Variance Equation C(4) -0.617559 0.154017 -4.009679 0.0001 C(5) 0.232701 0.041022 5.672543 0.0000 C(6) 0.076575 0.019992 3.830366 0.0001 C(7) 0.958289 0.012939 74.05972 0.0000 R-squared 0.012717 Mean dependent var 0.000423 Adjusted R-squared 0.004881 S.D. dependent var 0.005930 S.E. of regression 0.005916 Akaike info criterion -7.672488 Sum squared resid 0.026456 Schwarz criterion -7.629944 Log likelihood 2934.054 Hannan-Quinn criter. -7.656108 F-statistic 1.622938 Durbin-Watson stat 2.006895 Prob(F-statistic) 0.137822 Inverted AR Roots -.60 Inverted MA Roots -.71 Aim To study the leverage effect of news on exchange rate
All coefficients are positive
Bad news may not have a stronger impact than good news.
Last periods forecast has great impact
P-Garch Variable Coefficient Std. Error z-Statistic Prob. C 0.000249 0.000189 1.316457 0.1880 AR(1) -0.589688 0.142506 -4.137998 0.0000 MA(1) 0.707184 0.123404 5.730659 0.0000 Variance Equation C(4) 0.000156 0.000248 0.626489 0.5310 C(5) 0.122865 0.024157 5.086068 0.0000 C(6) -0.371455 0.087933 -4.224301 0.0000 C(7) 0.863090 0.025924 33.29256 0.0000 C(8) 1.073143 0.287283 3.735495 0.0002 R-squared 0.013432 Mean dependent var 0.000423 Adjusted R-squared 0.004285 S.D. dependent var 0.005930 S.E. of regression 0.005917 Akaike info criterion -7.671123 Sum squared resid 0.026437 Schwarz criterion -7.622502 Log likelihood 2934.534 Hannan-Quinn criter. -7.652403 F-statistic 1.468497 Durbin-Watson stat 2.013825 Prob(F-statistic) 0.175170 Inverted AR Roots -.59 Inverted MA Roots -.71 Aim To study the asymmetric impact of news on the volatility of exchange rate
Gamma is negative and significant
Bad news has a stronger impact on volatility of exchange rate than good news.
Component Garch All the coefficients are positive.
and are not significant.
This means if there is any bad news and the rupee depreciates the change would not be significantly higher than when a good news comes.