Predicting the next few numbers I have data that looks like this 

I just want to predict the rest of 2013. What is the best way of doing this. 
I used Excel's Forecast formula which uses linear regression. 
I only have minor/very little knowledge in stats.
 A: An appropriate model for this data is a simple AR(1).

The whole idea is to NOT assume a model NOR to ignore unusual values BUT to identify a minimally sufficient representation. The actual/fit/forecast graph is:

Even with 19 values it is sometimes possible to construct an equation which (reasonably) reflects the data and separates observations into signal and noise. This is a plot of the noise (residuals)

The residuals from an assumed exponential smoothing model are presented here for comparative analytics.

Notice the wide range for these residuals as compared to the empirically identified AR(1) model. The Excel Linear Regression model is also severly flawed as reflected by the following residual plot.

A: You can't really say much about this data without knowing where the data comes from. To forecast you need a model. All models come with assumptions and if you can't tell whether these assumptions are held then you won't be able to have any confidence in the predictions. For example, in linear regression, you assume that you data represents a process of the form: $$ y_i = \beta x_i + e_i$$, the $e$ accounting for the data not exactly fitting the straight line. It so happens that in order to have any confidence in your predictions from this linear model, every $e$ must be independent from the other. Can you guarantee that? Also, the $e$'s must be normally distributed, does the problem make that assumption sensible? There are other assumptions as well, but I hope you now understand why you can't just give a statistician data and say "please predict what happens next". 
