Forecasting Sample Model

2005-2007 Copyright. Dr. Johnathan Mun. All Rights Reserved.

Time-Series Forecasting
This sample model illustrates how to use Risk Simulator (Enterprise Simulator) for: 1. Running a Time-Series Forecast Model Background File Name: Time-Series Forecasting.xls The historical sales revenue data are located in the Time-Series Data worksheet. The data are quarterly sales revenue from Q1 2000 to Q4 2004. The data exhibits quarterly seasonality, which means that the seasonality is 4 (there are 4 quarters in 1 year or 1 cycle). Time-Series forecasting decomposes the historical data into the baseline, trend, and seasonality, if any. The models then applies an optimization procedure to find the alpha, beta, and gamma parameters for the baseline, trend, and seasonality coefficients and then recomposes them into a forecast. In other words, this methodology first applies a "backcast" to find the best-fitting model and best-fitting parameters of the model that minimizes forecast errors, and then proceeds to "forecast" the future based on the historical data that exist. This of course assumes that the same baseline growth, trend, and seasonality hold going forward. Even if they do not, say when there exists a structural shift (e.g., company goes global, has a merger, spin-off, and so forth), the baseline forecasts can be computed and then the required adjustments can be made to the forecasts. Procedure To run this model, simply: 1. Select the historical data (cells H11:H30). 2. Select Simulation l Forecasting l Time-Series Analysis. 3. Select Auto Model Selection, Forecast 4 Periods and Seasonality 4 Periods. Note that you can only select Create Simulation Assumptions if an existing Simulation Profile exists (if not, click on Simulation, New Simulation Profile, and then run the time-series forecast per the steps above but remember to check the Create Simulation Assumptions box). Model Results Analysis For your convenience, the analysis Report and Methodology sheets are included. A fitted chart and forecast values are provided in the report as well as the error measures and a statistical summary of the methodology. The Methodology sheet provides the statistical results from all 8 time-series methodologies. Refer to the User Manual or Modeling Risk (Wiley Finance, 2006) by Dr. Johnathan Mun for more detailed discussions of how to interpret the statistics and analysis results.

Disclaimer
DEVELOPER SPECIFICALLY DISCLAIMS ALL OTHER WARRANTIES, EXPRESS OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. As standard practice for software development and end-user applications, it is important for the Licensee to note that the valuation results attached herein are accurate to the software Developers best knowledge and are solely based on the information furnished by the Licensee or end-user. While the software Developer has used his best efforts in preparing this report, he makes no representations or warranties with respect to the accuracy or completeness of the contents of this model and specifically disclaims any implied warranties of merchantability of fitness for a particular purpose. The Licensee hereby agrees that the Developer is not held liable for any loss of profit or any other commercial damages, including, but not limited to, special, incidental, consequential or other damages. This model is only an illustration of using the software and in no way represents the correct and complete picture of an investor's investment, risk, and return profile. The user is advised to take great care in using and interpreting the model and its results. Model was developed by Dr. Johnathan Mun of www.realoptionsvaluation.com.

Copyright 2005-2007. Real Options Valuation, Inc. (www.realoptionsvaluation.com)

Historical Sales Revenues

Year 2000 2000 2000 2000 2001 2001 2001 2001 2002 2002 2002 2002 2003 2003 2003 2003 2004 2004 2004 2004 Quarter 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 Period 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Sales $684.20 $584.10 $765.40 $892.30 $885.40 $677.00 $1,006.60 $1,122.10 $1,163.40 $993.20 $1,312.50 $1,545.30 $1,596.20 $1,260.40 $1,735.20 $2,029.70 $2,107.80 $1,650.30 $2,304.40 $2,639.40

Copyright 2005-2007. Real Options Valuation, Inc. (www.realoptionsvaluation.com)

Holt-Winter's Multiplicative
Summary Statistics
Alpha, Beta, Gamma 0.00, 0.00, 0.00 0.10, 0.10, 0.10 0.20, 0.20, 0.20 0.30, 0.30, 0.30 0.40, 0.40, 0.40 0.50, 0.50, 0.50 RMSE 914.824 415.322 187.202 118.795 101.794 102.143 Alpha, Beta, Gamma 0.00, 0.00, 0.00 0.10, 0.10, 0.10 0.20, 0.20, 0.20 0.30, 0.30, 0.30 0.40, 0.40, 0.40 RMSE 914.824 415.322 187.202 118.795 101.794

The analysis was run with alpha = 0.2429, beta = 1.0000, gamma = 0.7797, and seasonality = 4

Time-Series Analysis Summary

When both seasonality and trend exist, more advanced models are required to decompose the data into their base elements: a base-case level (L) weighted by the alpha parameter; a trend component (b) weighted by the beta parameter; and a seasonality component (S) weighted by the gamma parameter. Several methods exist but the two most common are the HoltWinters' additive seasonality and Holt-Winters' multiplicative seasonality methods. In the Holt-Winter's additive model, the base case level, seasonality, and trend are added together to obtain the forecast fit. The best-fitting test for the moving average forecast uses the root mean squared errors (RMSE). The RMSE calculates the square root of the average squared deviations of the fitted values versus the actual data points. Mean Squared Error (MSE) is an absolute error measure that squares the errors (the difference between the actual historical data and the forecast-fitted data predicted by the model) to keep the positive and negative errors from canceling each other out. This measure also tends to exaggerate large errors by weighting the large errors more heavily than smaller errors by squaring them, which can help when comparing different time-series models. Root Mean Square Error (RMSE) is the square root of MSE and is the most popular error measure, also known as the quadratic loss function. RMSE can be defined as the average of the absolute values of the forecast errors and is highly appropriate when the cost of the forecast errors is proportional to the absolute size of the forecast error. The RMSE is used as the selection criteria for the best-fitting time-series model.

Mean Absolute Percentage Error (MAPE) is a relative error statistic measured as an average percent error of the historical data points and is most appropriate when the cost of the forecast error is more closely related to the percentage error than the numerical size of the error. Finally, an associated measure is the Theil's U statistic, which measures the naivety of the model's forecast. That is, if the Theil's U statistic is less than 1.0, then the forecast method used provides an estimate that is statistically better than guessing.

Period 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Forecast 21 Forecast 22 Forecast 23 Forecast 24

Actual 684.20 584.10 765.40 892.30 885.40 677.00 1006.60 1122.10 1163.40 993.20 1312.50 1545.30 1596.20 1260.40 1735.20 2029.70 2107.80 1650.30 2304.40 2639.40

Forecast Fit

684.20 667.55 935.45 1198.09 1112.48 887.95 1348.38 1546.53 1572.44 1299.20 1704.77 1976.23 2026.01 1637.28 2245.93 2643.09 2713.69 2114.79 2900.42 3293.81

Error Measurements RMSE 71.8132 MSE 5157.1348 MAD 53.4071 MAPE 4.50% Theil's U 0.3054

Copyright 2005-2007. Real Options Valuation, Inc. (www.realoptionsvaluation.com)

Methodology
The Best Model: Holt-Winter's Multiplicative Alpha 0.2429 Beta 1.0000 Gamma 0.7797 Seasons 4 RMSE 71.8132 MSE 5157.1348 MAD 53.4071 MAPE 4.50% Theil's U 0.3054 Second Best Model: Holt-Winter's Additive Alpha 0.2217 Beta 1.0000 Gamma 1.0000 Seasons 4 RMSE 92.4240 MSE 8542.1969 MAD 74.0883 MAPE 5.38% Theil's U 0.3424

Third Best Model: Seasonal Multiplicative Alpha 0.7195 Gamma 1.0000 Seasons 4 RMSE 152.3492 MSE 23210.2897 MAD 126.6442 MAPE 9.01% Theil's U 0.4908

Forth Best Model: Seasonal Additive Alpha 0.6882 Gamma 1.0000 Seasons 4 RMSE 182.4348 MSE 33282.4687 MAD 161.1926 MAPE 10.62% Theil's U 0.5607

Fifth Best Model: Double Exponential Smoothing Alpha 0.1160 Beta 1.0000 RMSE 223.6280 MSE 50009.4736 MAD 197.9173 MAPE 15.09% Theil's U 0.7313

Sixth Best Model: Double Moving Average Periodicity 4 RMSE 272.7573 MSE 74396.5245 MAD 249.0010 MAPE 14.40% Theil's U 0.7248

Seventh Best Model: Single Exponential Smoothing Alpha 0.7948 RMSE 287.1776 MSE 82470.9621 MAD 249.2510 MAPE 18.16% Theil's U 0.9561

Eighth Best Model: Single Moving Average Periodicity 2 RMSE 318.8152 MSE 101643.1317 MAD 280.3222 MAPE 19.55% Theil's U 0.9623

Copyright 2005-2007. Real Options Valuation, Inc. (www.realoptionsvaluation.com)