This can help: http://www.bishophill.com/admin/sidebar_images/1741759940_test.pdf

I would not reccommend using ARIMA modelling with only 24 observations. In fact only 23 if you need to difference the data to remove the trend.  I suggest you difference the data ( in logs) and look at the acf and pacf. If the suggested model is MA1, ghen it is simply trend + random. You might as well fit a linear/log trend using OLS

A sample size of 24 is too little for time series analysis. The realization should at least be of length 50 for any meaningful statistical inference. Check Box and jenkins (1976) for instance.

A series of 24 observations considers a very short series.  A simple simulation study were conducted about this topic as explained in the following link

http://robjhyndman.com/hyndsight/short-time-series/

You might find the link below useful

https://www.researchgate.net/post/What_should_be_the_minimum_number_of_observations_for_a_time_series_model