# L-48 NG weekly storage level

### Introduction

The Energy Information Administration (EIA) publishes a weekly storage report for natural gas.The NG and power market participants monitor those reports closely, and, strive to forecast them; especially the ending season levels, as an indicator of where of gas scarcity, and thus price levels.

### Objective

For our purpose, we will model the underground storage level in the lower 48 states in US, and forecast the levels for different intervals.

### Analysis

Examining the graph above suggests that our process exhibits seasonality (52 weeks), and seems weakly stationary (i.e. stable variance over time).

Using TSACF and TSPACF functions, we compute the ACF and PACF for different lags. The correlgram for our data is shown below:

 ACF PACF

The ACF and PACF correlgrams above suggest that we have an auto-regressive process of second order AR(2). The PACF for 3rd and fourth lag are significant non-zero, but relatively small.

For AR model, we ran Excel multiple regression routine (part of Excel Analysis pack) to estimate the model parameters. X1 is the first lagged time series and X2 is second lagged time series.

$$y_{t}= \theta_{0} + \theta_{1}y_{t-1} + \theta_{2} y_{t-2} + \epsilon_{t}$$

$$\epsilon_{t}\sim N(0,\sigma^2)$$

The regression output shows all parameters to be significant and, the model has an Adjusted R2 of 99%.

Next, we use the model above to forecast storage levels for the coming weeks/months;

The model error is 38 Bcf, so our confidence interval (for 5% significance) is +/- 112 Bcf. Although this model has a great R2, the residuals error is huge.

Using the data as of the week of March 13th, 2009. The model suggests that storage will be at its lowest level (1622 Bcf) by end April 3rd week.