# There to develop a model to forecast conditional heteroskedasticity.?

There are eight categories of data released daily for the eight major currencies or countries thatare closely followed in Forex market. The currencies are as follow: USD, EUR, GBP, JPY, CHF,CAD, AUD, NZD? As the general rule because US dollars is in more than 90% of trades, US economic news tend tohave the most impact on the market.? Asset holders are interested in the volatility of returns over the holding period, not over somehistorical period. This forward-looking view of risk means that it is important to be able toestimate and forecast the risk associated with holding a particular asset. To model and forecastvolatility I need to develop a model to forecast conditional heteroskedasticity.? The unconditional variance (the long-run forecast of the variance) would be unimportant if Iplan to buy the asset at t and sell it at t+1.? Conditional forecast means I have all the information up to today and I want to forecasttomorrow.? To build the model I use Internet search queries for the “keywords in News” related to eachcurrency. These queries can be obtained from google trends.? I use ARCH model since most of the models related to volatility estimation of the financialmarkets are ARCH and GARCH models.? ARCH and GARCH are based on stationary data. So I need to make sure that my news Googletrends are all stationary.? I also do the LM test for ARCH model to check whether conditional heteroskedasticity exists.? Y t+1 =e t+1 x t. Where Y t+1 is the variable of interest that is volatility, e t+1 is the white-noise disturbanceterm with the variance of sigma squared and x t is an independent variable that can be observedat period t which is news Google trend. I will have 8 independent variables. Here the realizationsof the x t sequence are not equal, so the variance of Y t+1 conditional on the observable value of x t .? Since I can observe x t at time period t, I can form the variance of Y t+1 conditionally on the realizedvalue of x t .? One simple strategy is to forecast the conditional variance as an AR(q) process using squares ofthe estimated residuals.e t 2 = a 0 + a 1 e 2 t-1 + a 2 e 2 t-1 +…..+ a q e 2 t-q + v tIs an auto regressive process which is the ARCH (Autoregressive Conditional Heteroskedastic)model, v t is the white-noise process. I can use this equation to forecast conditional variance att+1 asE t e 2 t+1 = a 0 + a 1 e 2 t + a 2 e 2 t-1 +…..+ a q e 2 t+1-q? This model is the linear specification which is not the most convenient.? The conditional heteroskedasticity in { e t } will result in {y t } being Heteroskedastic itself. So theARCH model is able to capture periods of tranquility and volatility in the {y t } series.