Table 9. Granger causality test.

Dataset	lag(days)	Direction			Avg reduction in RSS
Q (100 tickers)	1	Q T
Q (100 tickers)	1	T Q
U (100 tickers)	1	U T
U (100 tickers)	1	T U
Q (100 tickers)	2	Q T
Q (100 tickers)	2	T Q
U (100 tickers)	2	U T
U (100 tickers)	2	T U
Q (87 tickers)	1	Q T
Q (87 tickers)	1	T Q
U (87 tickers)	1	U T
U (87 tickers)	1	T U
Q (87 tickers)	2	Q T
Q (87 tickers)	2	T Q
U (87 tickers)	2	U T
U (87 tickers)	2	T U

Adding information about yesterday’s query volume reduces the average prediction error (in an autoregressive model) for today’s trade volume by about , and for half of the companies the reduction is statistically significant at .