Issues in Revenue Forecasting

Transcription

1 Issues in Revenue Forecasting Rich Simons Itron, Inc Energy Forecasting Week Las Vegas, Nevada Forecasters Forum/EFG Meeting Forecasters Forum/EFG Meeting April 29 30, 2010

2 Linking the Sales Forecast with Revenue Ultimately the purpose of the budget sales forecast is to generate revenue estimates and forecasts > Financial i planning and budgeting > Evaluating new tariffs > Revenue requirements calculation > Variance analysis and financial reporting There is a trend toward more complex pricing strategies which complicates the translation of billed month sales forecasts to revenue forecasts.

3 Revenue Forecasting Can be Complex Sales and customer forecasts are generally developed at the customer class level while revenues are often required at the rate and rate block level Revenue forecasts are typically required on a calendar month basis, while rate data is on a billing month basis Rate classes do not necessarily line up with customer classes > Rates can span multiple revenue classes Billing demand (kw), Reactive Power (kvar), demand ratchets, and load factor blocks can be difficult to model > Difficult to define for a calendar month vs. revenue month basis > Billing demand is often calculated as the sum of the maximum demands from multiple customers, which occurred at different times during the billing month

4 The Low Road Average $/kwh > Calendarize the forecast > Calculate a $/kwh that accounts for the various components: Blocks Billing Demands > Multiply the forecast by an average price: kwh x $/kwh > Get out of town Benefits > Fast to develop > Easy to understand and explain Downsides > Not enough billing detail to support various functions Revenue variance analysis Cost of service studies Pricing analysis

5 The High Road Forecast the tariff components and price them out Calculate historical rate shares Share = rcm r = Rate Sales rm Sales cm c = Class m = Month Res Standard Rate Share Calculate historical blocks shares for each rate Sl Sales Share = brm brm Sales rm Block 1 shares are highest in May and October when less energy gets into the upper blocks b = Block Forecast the shares Share out the forecast to rates and block Multiply l by prices Block 1 Share

6 Bumps in the High Road Rate and block data may not be readily accessible Rt Rate dt data can be extremely noisy > Billing adjustments > Billing errors > Migration Once you do obtain the data, it can be difficult to manage it Rate data may not line up exactly with the reported class data because some adjustments may be made at the class level Rate definitions may not align easily with class definitions

7 Using MetrixND for Revenue Forecasting MetrixND provides a powerful tool for modeling billing determinants and revenue > Build complex determinants dt t with Transformation Tables Tbl > Can graphically view billing determinant trends Share of on peak usage, Billing demand Load factors > Link to historical determinant data and tariff level rate forecasts in external databases > Link to MetrixND budget sales models > Generate determinant and revenue reports

8 Calculate Rate Class Sales Ui Using MtiND MetrixND 1. Estimate the proportion of class sales accounted for by each rate 2. Apply these proportions to the class forecast to generate an estimate of the underlying sales 3. Forecast the rate sales using these allocated sales forecasts as drivers 1 2 Sum to 1.0 3

9 Ideal Integration with Forecast Manager Streamlines data acquisition Calculates: > Daily, calendar month, billing month, unbilled period weather concepts > Average billing days and unbilled days > Normal weather concepts Integrates with MetrixND and extracts: > Normalized sales > Unbilled energy estimates > Calendar month estimates > Forecasts Tracks and compares forecast results Archives forecast vintagesorscenarios scenarios

10 ForecastManager

11 Revenue Integration with FM FM Database Historical Rate Data 1 Forecasts Price Assumptions 4 2 Historical Rate Data 3 Class Energy Forecasts Weather Data Price Assumptions Rate Forecasts: kwh and kw Customers Dollars

12 Residential Rate Structure 1 Class Residential 2 Rate Standard Electric Water Heat Electric Heat & Load Management 3 Block Less than 500 Kwh Greater than 500 Kwh Less than 500 Kwh Greater than 500 Kwh Less than 500 Kwh Greater than 500 Kwh

13 Residential Rate Shares Regression Model: Share = F(Trend, Months, MA1) A. Standard Share: Standard/Residential A History Forecast Residential B. Electric Water Heat Share: 1 (A+C) B C. Electric Heat & Load Management Share: EHLM/Residential C Regression Model: Share = F(Trend, Months, MA1)

14 Standard Block Shares A. Less than 500 Kwh: LT500Kwh/StandardKwh A Standard Kwh B. Greater than 500 Kwh: 1 A B

15 Electric Water Heat Block Shares A. Less than 500 Kwh: 1 - B A Electric Water Heat Kwh B. Greater than 500 Kwh: GT500Kwh/EWHKwh B

16 Transformation Tables Total Residential Residential Standard Rate Residential Electric Water Heating Rate One Transform Table for Total Class and each Rate s Share to Total. This way the variables have analogous names, while the Table name describes the rate. Residential Electric Heat

17 Residential Standard Share Model Simple trend and seasonal models

18 Electric Heat & Load Management Block Shares Electric Heat & Load Management Kwh A. Less than 500 Kwh: 1 - B B. Greater than 500 Kwh: GT500Kwh/EHLMKwh A B

19 Residential Customer Shares Residential Customers A. Standard Customer Share: Standard/Residential A B. Electric Water Heat Customer Share: EWH/Residential B C. Electric Heat & Load Management Customer Share: 1 (A+B) C Project customer shares with simple trend models

20 Commercial Customer Shares Commercial C i l Customers A. B. General Primary Customer Share General Service Customer Share A B C. General Service Sub-transmission Customer Share C D. General Transmission Customer Share D Commercial customer shares tend to be more erratic than the residential. E. Outdoor Lighting Customer Share E

21 Cycle Geometry A B B C C Revenue Month Calendar Month: Model Simulations Unbilled Corner: Model Simulations

22 Unbilled Simulation Simulate with Unbilled CDD & HDD Unbilled Days Unbilled Sales per Day Regression Model Cycle Sales per Day = F(Trend, CycCDD, CycHDD, Binaries) Simulation Unbilled Sales per Day = F(Trend, UnbCDD, UnbHDD, Binaries) Transformation Unbilled Sales = Unbilled Sales per Day X Unbilled Days

23 Sharing the Results Revenue Month Use Transformation Tbl Tables to apply the Model dl results and Simulations. Calendar Month Unbilled

24 Excel Spreadsheet with VBA Uses VBA to open MetrixND and to extract results from reports objects MetrixND Report Formatted Excel Report

25 Summary MetrixND is a powerful tool for developing billing determinant and revenue forecasts > Specify and view determinant calculations clearly > Visualize calculation results > Build simple regression orsmoothing models toprojectbilling determinants > Link to budget sales forecast models and external databases MtiND MetrixND models dlcan be linked with ForecastManager to provide a means for managing, archiving, and tracking the budget and revenue forecast But be aware: Forecasting revenues at the rate and block level implies a certain level of precision that may not be there.