How is the normal temperature calculated?

Normals are based on a 10 year climatology utilizing the NOAA CLIM84 standards which apply a cubic spline to the monthly averages to create daily normal temperatures. These are then applied to the 11 demand weighted UK LDZ sites and aggregated creating a single UK Daily Normal temperature.

How is performance monitored?

Skill scoring is a built-in performance analysis of FAST Trader. Metra will monitor and record the MAE and bias of our own propietary forecast vs ECMWF & GFS. You will find that Metra's forecast has less bias and more accuracy than the current and widely used models available. With this information, trading decisions based on weather forecasts will be better informed and can be executed with greater confidence.

Comparing the Accuracy of Weather Forecasts

Definitions:
MAE - Mean Absolute Error - Is a measure of how close the temperature forecasts are to the actual temperatures.

RMS/MSE - Root Mean Square/Mean Square Error - Is the square root of the mean of the squares of the forecast values. The squaring function penalizes the temperature errors at a non-linear rate thus making larger errors more prominent.

Bias - Is a measure of the forecast system over predicting (too warm) or under predicting (too cold) temperatures.

Skill Testing:
Example 1: Using Bias

When assessing weather forecasts, not only is the forecast important but also the CONFIDENCE and RISK surrounding the forecast. The bias allows you to see where some of the directional RISK to the natural gas price might lie. Are the weather models such as the ECMWF or GFS too warm and the RISK to the price is downward? Or, are the numerical weather models such as the ECMWF or GFS too cold and RISK to the price is upward? If you subscribe to Fast Trader you will not only be able to see the bias of the ECMWF or GFS but you will be able to compare those results against Metra's own proprietary forecast to help you gain CONFIDENCE on how much to weighting should be placed on the current numerical weather models, such as the ECMWF and GFS.

Example 2: Using Bias

If the market is pricing in too much cold in the winter then the prices are too high and if they are not factoring in enough cold in the winter then the prices will be too low. By you running your own running supply and demand models you will see:

• if the natural gas in storage is near expectations perhaps due to very little temperature bias;
• if the natural gas in storage is below expectations perhaps due to the weather forecasts coming in colder than expected (cold bias) and thus demand has been more than anticipated;
• if the natural gas in storage is above expectations perhaps due to the weather forecasts have been coming in warmer than expected (warm bias) and thus demand has been less than anticipated.

The prices in the market will need to shift upward or downward to reflect the fair value for the natural gas available. When comparing the ECMWF and GFS numerical weather model against Metra1s's own proprietary forecast you can perhaps gain a clearer picture on which way the market prices need to adjust.

Examples 3: Using RMS

Although Metra's proprietary forecast may not always be in first place for every time period you can see how Metra's forecast compares to the other numerical weather models (ECMWF and GFS). For example, if Metra's forecast and the ECMWF deterministic model run are performing the best then you can discount the ECMWF ensemble, GFS ensemble and GFS deterministic runs.  By lining up more than one datapoint you can increase your CONFIDENCE in Metra's own proprietary forecast and perhaps a particular numerical weather model, i.e., the ECMWF or GFS for a specific time period.

How to Get the Most out of the Analysis Tool

This powerful analysis tool allows the user to the track the trends in the weather forecast over the past 1 day (GFS) to 2 days (ECMWF). Below are some examples of how this tool can help you assess the weather forecast:

• Are the numerical weather forecasts trending warmer or colder? Is that in-line with market expectations? If not, there could be a price adjustment.
• Are all the weather model forecasts showing the same 'shape' and hence creating a higher CONFIDENCE in the forecast or are the forecasts diverging thus indicating more RISK and less CONFIDENCE in the forecast?
• Was there a big change in the forecasts? Did the weather models analyze more high pressure and indicate a much colder and drier forecast? Did the weather models analyze more low pressure indicate a warmer, wetter and windier forecast? Should you believe the change? Are the subsequent weather forecasts reinforcing the initially indicated pattern change?
• Did Metra's proprietary forecast indicate either a warmer or colder change before the next model run of the GFS or ECMWF thus allowing you to amongst the first to react in the market?

Five day average analysis tools

The tool allows the user to choose and plot the five day average of the observations, five day average of the climatology and the five day average of the GFS and ECMWF deterministic and ensembles runs over the current 24 hour period and then overlay it one chart.

Running five day average analysis tool.

The single horizontal line represents the five day temperature average of the 11 UK LDZ sites for the preference that is picked - observations, climatology and/or weather model data.

Work week five day average analysis tool.

The same concept as the running five day average analysis tool, but instead calculated for work week.

How can I use the five day average tool?

The user can plot the previous observations, climatology and the weather model output to determine if there are discernible trends in the data. For example, if the current model run came in much colder, on average, than the previous models runs for the 11 to 15 day period you can watch the trend in the subsequent weather models to determine if it was an outlier or indicated a trend thus allowing for forecast adjustments.

How are the box and whiskers calculated?

• the graphs are made utilizing all the max and min from each day period contained in the five day period;
• from top of top thin whisker to bottom of bottom thin whisker represents 100% of the values for that five day period;
• from top of top thick whisker to bottom of bottom thick whisker represents 80% of the values for that five day period;
• from top of the box to bottom of the box represents 40% of the values for that five day period;
• the thick horizontal line represents the average of all temperatures for that five day period;

How can you use the box and whiskers?

The important pieces of information that can be used from using a box and whiskers delta are visualizing the trend and the changes in the distribution of the forecast over time. Since the forecast is a distribution, not a single point, the user can use how the distribution is changing over time to gain CONFIDENCE and identify the RISKS in the forecast.

The user can ask the following questions to help gain that confidence and identify those risks in the forecasts:

• Are the forecasts trending warmer or colder over time?
• Are the forecasts trending above or below normal?
• Is the area in the box representing 80% of the distribution increasing or decreasing?
• Is the area in the box representing 40% of the distribution increasing or decreasing?
• How is the shape of the forecast distribution changing? Is there more of warm tail or cold tail?
• Are the outliers mostly to the warm end or the cold end?
• How does the trend in the deltas compare between the GFS, ECMWF and Metra?
• How does the trend in the forecast compare to the current conditions? Colder? Warmer?
• Are the forecast deltas (ECMWF, GFS, Metra) converging or are the diverging?