Ok, here some results.
To determine if (tick) volume of current candles have any predictive power, I first estimated up and down tick volume using my formula discussed above. For each M1 candle of 100001 EURUSD bars I calculated then the volume difference = up volume – down volume for each candle. Then the idea was to search for highest correlations between this volume difference of current candle on one hand, and the prize range (calculated as prize change = close – open) for one of the subsequent or preceding candles. This is exactly, what the cross correlation function does. The first attachment below shows exactly this cross correlation.
The highest correlation can naturally be seen for a time lag of 0. The second highest correlation we obtain for a time shift of +1. Althought it is significant from 0, the correlation between voldiff and close-open is not very high (r ~ -0.024). Additionally, this correlation is no longer positive. It can be interpret that in most cases we obtain a big volume difference (either much more up than down or more down than up volume), we can expect a move of next candle in opposite direction.
Therefor, I’m absolutely not sure if volume is really leading prize. It think it is correlated with prize of current candle and nothing more. This conclusion can also be drawn from the regression of lag = 0 (where we assume that volume is independent and prize change depends on it) which is attached as second image. But this doesn’t mean that we can’t use volume for trading. Maybe it may give good signals during extreme market situations as @gg53 mentioned. Perhaps it can be misleading to apply findings under such market situations on normal market situations. Nevertheless, that regression analysis revealed that the association between range and volume is highly significant p < 2e-16 and that about 78.4 % of the variance of the range (close – open) can be explained by current candle’s volume (R^2 = 0.7844).
That’s the conclusions from the 1 bar analysis. A rather interesting result was obtained when looking on the ccf of volume difference of current candle and the range change over the next 5 bars (close-open = range5). It’s shown as third attachment. Here we see that range5 is negatively correlated to volume difference of bar i. And this is not only true for lag = 1 but also for lag = 2 and lag = 3. The conclusion from this finding is that volume can be leading. I would expect that especially under high volume differences we can expect a move to the opposite direction over the next 7 bars.
The next step of analysis should be to evaluate when a volume difference is high enough to trust the move to opposite direction. Any questions, suggestions, comments? Don’t hesitate …