As US market data keeps markedly improving the US dollar is poised to keep getting stronger as the US dollar keeps its safe-haven status in the midst of high US Treasury yields and a slowing Chinese economy. This in turn is going to prove difficult for emerging market currencies and BRICS hopefuls to recover their currency’s recent losses to the US dollar. Given the current scenario it would be the prime time to bet the US dollar would appreciate further in the coming months. However, like the stock market, FX markets aren’t that forgiving either. From knowing strong currency pairs like USD/EUR to monitoring activities of the Chinese Yuan and Indonesian Rupiah as used in international commerce, there is a gamut of complexities surrounding FX trading.
To start building the necessary know-how to participate in FX trading is to familiarize oneself with the currency of choice and the intricacies of its data, specifically a currency’s time series data. In financial time series data, much like stock returns, foreign exchange (FX) rates may appear uncorrelated over time but are not interdependent. However, to conduct meaningful forecasting with such data, it’s crucial to consider three key characteristics: kurtosis, skewness, and volatility.
Numerous studies have indicated that FX time series data exhibit thicker tails compared to a normal distribution, resulting in excess kurtosis, often referred to as leptokurtic. This implies that the data, which would otherwise follow a normal distribution, has relatively few observations at the extremes and an abundance of observations near the mean. Moreover, it’s suggested that an exceptionally large positive return (indicating currency depreciation relative to a reference currency) will lead to extended tails in the distribution on the following day. Consequently, in the short term, extremes tend to be followed by other extremes before a market reversion occurs after the second day.
Another crucial characteristic in FX data that significantly impacts forecasting is skewness. Numerous studies have demonstrated the presence of skewness, although it is less common due to the two-sided nature of FX but not impossible. Notably in using high-frequency intraday data for FX, it has been observed that employing model-free estimates (to minimize measurement errors, achieved by estimating daily volatility through the summation of high-frequency intraday squared returns) reveals skewness, particularly skewness to the right.
For volatility, comparisons of FX data reveal highly correlated volatility movements. Furthermore, it has been found that the correlation between exchange rates increases with volatility. Their results reinforce the evidence of significant volatility clustering effects in daily returns, with monthly realized volatilities also displaying high persistence. Consequently, foreign exchange rates demonstrate leptokurtosis, right-skewness, and pronounced volatility clustering effects.
Now putting those three characteristics present in FX data and applying its understanding to the movements of FX data through time are the necessary ingredients to employ the use of the usual forecasting tools applied by most quants today for FX trading. These include the GARCH family of models that employ the use of FX volatility to create forecasts based on assumed scenarios usually backed by qualitative methods to ensure their forecast models are fit for purpose.