Volume data alone does n't make sense. Please see the price change and volume change together with the specific dates, you will get insightful  results. Theoretically, it is said that any price action should be backed by volumes. Volume trends become strong at support and resistance level also.

Thanks. For a moment, lets focus on Volume not price for the sake of discussion.

Since the volume byitself is not a time-series data, then even if you take ROC (rate of change) in volume then again the same "non time series" properties will refflect in Volume ROC. So what's the point of take any Moving average of ROC of Volume?

Volume is a time series data if you take the volume with the date of trading and if rate of change is used, you may get some interesting results. Again, You need to be very careful about the interest variable in the market. 

Is volume of trades a time series datum?  Events are described by their properties.  The volume of occurrences described as "trade that occurred at time t" is an event property, and a series of such event properties tagged to time of occurrence is a time series.  For these reasons, I agree with Ranajee.  Independently of whether volume is a time series datum, Ankur's argument seems to be that historical moving averages of volume are not comparable with current volumes because of high standard deviation in this property.  But this claim sets up an incomplete and unresolvable dichotomy.  What is the threshold?  How high a standard deviation makes historical volume not comparable to current volume? 

The moving average, whether it be of price, volume, or something else, shows a trend or the absence thereof.  That is the whole idea of regression or least squares. Historical trading volumes might better be compared with current trading volumes than with the heights or weights of the traders.  If, in Ankur's words, "the moving averages of Volumes for Any Time Period is not the right metric to compare with the Target day's Volumes", then what is?  It is like saying that the historical moving average of the Richter number of earthquakes centered at a given location (where earthquake Richter numbers show a high variance) cannot be meaningfully compared with the Richter number of earthquakes in the current observation interval. This is not to say that the moving average of trading volume might not be more or less meaningful depending on the observational frequency or inter-observational interval. Care in selecting these test parameters is warranted.

@RanaJee:

When you say "interest variable", are you referring to the sentiment, corporate disclosure, Index movement, industry movement, Price Impact (all these at the Event date/ volume Noise date)

Is that you meant with interest variable? Also, is there any way to Quantify it or it should be judgmental?

Attaching a paper by Jonathan M. Karpoff for your reference. 

Not yet seeing Ranajee's answer to Ankur's questions of 11 hours ago, I'm thinking that the "interest variable" may refer to the time-value of monetary units (e.g. dollars, euros, etc.) as dictated by central bankers.  Certainly, this is a meaningful property and adjunct-variable of the price, stated in those same units, of an equity distributed over time -- as would be also the geometric mean of price for an aggregate of monetary commodities (fungible metals).

Moving average is a filter which filters out random effect. It is used in electric engineering in this way. If you understand trend as a constant linear increase/decrease you are maybe wrong. Moving average show the main course of events including cycles for example. If you use regression analysis to fit a curve then there is no significant difference in the qualities and in the fitted function if it is fitted to the original data or tho moving average. However the shape of moving average may give idea which function should be fitted.

@Albert:

I am absolute certain @Ranajee did'nt mean "interest variable" as time-value of money or any discounting factor.

Also, with lot of self discussion I believe, it's scope  is sentiment, corporate disclosure, Index movement, industry movement, Price Impact (all these at the Event date/ volume Noise date) and few more variables that has direct impact on Liquidity. 

About Quantifying: I would believe it's best done Judgmentally. (I am doing backtesting to see if at all this can be Modeled with some accuracy. But again, I am a discretionary kind of Person (not a wholesale analytics guy) by personality. Modelling everything only makes things over complicated and Sensitive to independent variable.

Ankur, I understand.  You are the researcher and it is entirely your judgment what variables are significant.  That said, I would emphasize that the time value of the monetary unit becomes a more and more significant factor in interpreting equities price trends as the duration of your observation covers more time or as the rate of depreciation of the monetary unit increases.  In the U.S. I have sometimes seen asset price graphs over many decades and find them relatively useless unless converted to normalized "constant" dollars.  A unit is, after all, only a unit if it is invariant with time.

Bela Vizvari has made a series of excellent points.  In my earlier posting I intentionally avoided mentioning curve fitting, not to complicate matters.  But there is indeed a limit to what one can learn from linear regression when dealing with significant non-linearities.  I have found it useful to test for or at least estimate non-linearity as a precursor to using autoregressive moving average to interpret or project the strength of a trend in dependent variable, and perform a curve fit if appropriate.

Ankur,

I agree with Albert that we are dealing with a time serie. This time serie may be decomposed in the sum of a trend component and quick fluctuations (with a random/fractal nature). As I understand correctly, you feel like a moving average is not helpfull to study volume. I think we have to re-formulate this a bit. Suppose we can strip the time serie from it's random fluctuations. Maybe then the moving average provides you with good insights. In other words, it seems the time serie is a tough nut to crack in terms of revealing it's trending behavior. I recommend you to read this paper:

A mathematical proof of the existence of trends in financial time series

Michel FLIESS & C´edric JOIN  

An excellent paper written by two researchers with a fresh look at financial time series behavior.

Henk-Jan Ramaker, is this an archival paper?  It did not come up immediately in a multi-database search.   A more complete citation would be helpful  -- thanks in advance!

Hi Albert, you can find using the link below.

https://arxiv.org/abs/0901.1945