Autocorrelation:

If we denote the error term as u_i for the i^th observation, then we assumed that:

where cov(x,y)  is the covariance between the two variables  x and y'.  However if this assumption is  violated then there exists autocorrelation.  that is,

There are typically two types of data used for empirical studies: cross section and time series. In  the first case data on various variables are collected across  a number of households or firms at a given point  in  time. Examples in India  include the Economic Census, and  the Census of  India. This  type of data can be used to test for any relationship between variables across firms or households (depending on the sampling unit). For instance, are bigger firms  (in  terms of revenue) more productive (Per unit of labour) is  a question that can be answered using the' Economic Census data. Data may also  be collected over time for a set of variables  for  the same sampling unit (usually at country or state level, may also be for households or firms). This type of data is known as  time series data. This  allows us to check  for any relationship between  the variables for the same unit, over  time. For  instance, if aggregate consumption  and income data are collected at  the country level, then we can  estimate the consumption function, that is, the  relationship between  income  and consumption. As you know from macroeconomic analysis, here we can answer questions such as: Does consumption increases with income and if so by what amount? Another form of  data that has gained prominence in  recent  times  is when time series  data  is available for multiple units, then we say this is pooled (time series and cross-section data), also called panel data.