In this paper, we have examined the short-run movement of the won/dollar exchange rate using information obtained from the inter-bank market in Korea. First, we constructed the hourly measure of excess demand for dollar and used it as a proxy for the trading pattern of market participants. To construct this time series, we relied on the bid and ask won/dollar exchange rates collected on the two-minute interval. We then estimated the structural VAR model consisting of the actually observed won/dollar exchange rate and the proxied trading pattern of market participants to see if private information, as opposed to public information, is relevant for explaining the hourly movement of the won/dollar exchange rate. Private information is found to account for more 30% of hourly variations of the won/dollar exchange rate. Next, we constructed the trading pattern of market participants on a daily basis using the same data set employed to build the hourly measure. We then examined whether private information is useful for predicting the daily won/dollar exchange rate movement. We found that the forecast model using both private and public information reduces out-of-sample forecast errors of an alternative model relying only on public information by 20~25%. Also, the out-of-sample forecast of the model using both private and public information is found to be more accurate than the random walk model.