Mixed-data sampling (MIDAS) is a family of regression models that relate a variable observed at a low frequency to explanatory variables observed at higher frequencies, without first averaging the frequent data away. It was introduced in the mid-2000s by the economist Eric Ghysels and co-authors, initially for financial volatility, and is now a standard tool in econometrics.
A typical use is nowcasting quarterly GDP from monthly surveys, industrial output or daily market prices. Rather than giving each high-frequency observation its own coefficient, which would require too many parameters, MIDAS uses a small weighting function, such as an exponential Almon or beta polynomial, to decide how much each lag counts. This keeps the model compact while letting recent data matter more. Central banks, including the Bank of England, use MIDAS and related time series forecasting methods to track the economy before official figures are published.
Ref 1100001543