Converting monthly stock data into quarterly data for DSGE models: taking mean of monthly data or taking the last observation

Dear Johannes,
Thank you very much for your helpful guidance, I am grateful.
I am preparing data like exchange rate and stock prices, both of which are of monthly frequency, and I need to change them into quarterly data. For stock variables/non-flow data, should I taking the average of each month for the quarterly data? Or should I take the last monthly observation as the quarterly data? I have read the literature, regarding monthly stock prices and exchange rates, some researchers take mean value of monthly data into quarterly data, however, some researchers say that only the last observation of monthly data should be taken as the quarterly observations, because averaging monthly stock prices into quarterly version introduces spurious serial correlation in returns (the “Working effect”), and they say I should calculate returns using end of quarter natural log prices. However, in the attached paper, monthly stock prices are averaged to obtain quarterly series, please refer to the PDF attachment on page 1710 (I have highlighted them in yellow).DSGE1.pdf (674.7 KB)

I have another question, do most DSGE models have variables not percentualized? Because on the same page of the PDF document, it says ‘all variables are not percentualized.’
Thank you very much and look forward to hearing from you.
Best regards,
Jesse

  1. You need to decide what reflects the concept in your model best. Most of the times, you are considering the stock price at the end of the quarter reflecting the dividends paid in the present period. In that case, the end of quarter value should be most appropriate.
  2. I don’t get the part about being “percentualized”. In equation (49) of that paper, there is clearly a log-difference in the transformation. Or do the authors simply want to state that they are not taking the log twice?
1 Like

Dear Johannes,
Does ‘Not percentualized’ here mean ‘not multiplied by 100’? Because I find the paper log-differences some data.
And is it ok to not multiplied by 100 after log-differencing the data? Given that measurement equations do not multiply log-differenced state variables by 100 to obtain observable varaibles.
Thank you very much and look forward to hearing from you.
Best regards,
Jesse

That could be. The default is not multiplying with 100. There is no need to do so. But sometimes it is more convenient.