SQLRollingStats

Easy to Use
All SQLRollingStats functions are implemented as types: you need only declare a variable of the appropriate type, assign it a starting value (usually '0') and then call its Result() method in your queries using the syntax @variable.Result(),  specifying the input values and a row-number (since almost all forms of rolling statistics are only valid when processed in a given order). T-SQL's built-in ROW_NUMBER() function will usually suffice for the purpose - see the examples given in the Product Documentation for more details.

Support for Tiling
Just as the built-in T-SQL ranking functions such as ROW_NUMBER() support tiling (the ability to group results into distinct blocks and re-start the calculation from scratch at the start of each block) using the PARTITION BY clause, so do the SQLRollingStats functions - so you can in a single query calculate rolling statistics across a number of different domains.

Outstanding Performance and Numerical Accuracy
The algorithms behind the SQLRollingStats functions use industry-proven, best-in-class techniques devised by some of the world's leading numerical computing experts to maximise performance and accuracy even for fairly extreme input values. Careful heuristics and edge-case checks through the code ensure that problems caused by the limitations of floating-point arithmetic on digital computers (such as cancellation errors) are kept to an absolute mininum, and a broad range of special-purpose, high-performance numerical computing approaches are used in place of the slow, expensive computations that would result from 'naive' implementations of the underlying mathematics.

SQLRollingStats Features

• 30+ scalar functions that return results based on data from current row and prior rows in the same query.All functions support a rolling window that discards older values, or can be cumulative (no window)
• All functions support tiling (accumulation of data restarts when input row number restarts from 1)
• All functions provide checking & enforcement of row-processing order using row number inputs
• Sum and product
• Maximum, minimum, range and mid-range
• First (for a window size of n will return input from n rows ago) and Previous (value from prior row)
• Average (mean), sample & population variance, sample & population standard deviation
• Geometric mean, harmonic mean, quadratic mean
• Weighted average, sample & population variance, sample & population standard deviations. 
• Exponentially weighted average, variance and standard deviations
• Weighted power mean (generalized mean)
• Median, quartile, inter-quartile range, tri-mean
• Percentile & percentile rank