140 likes | 298 Views
Demand Management and Forecasting. Types of Forecasts. Qualitative Time Series Causal Relationships Simulation. Qualitative Methods. Grass Roots. Executive Judgment. Prediction Markets. Qualitative Methods. Market Research. Historical Analogy. Delphi Method. Panel Consensus.
E N D
Types of Forecasts • Qualitative • Time Series • Causal Relationships • Simulation
Qualitative Methods Grass Roots Executive Judgment Prediction Markets Qualitative Methods Market Research Historical Analogy Delphi Method Panel Consensus
Prediction Markets http://abcnews.go.com/video/playerIndex?id=4826867 http://www.intrade.com/
Quantitative Approaches • Naïve (time series) • Moving Averages (time series) • Exponential Smoothing (time series) • Trend Projection (time series) • Linear Regression (causal)
Naïve Method • This period’s forecast = Last period’s observation • Crude but effective • August sales = 1000; September sales = ?? • 1000!
Moving Averages • This period’s forecast = Average of past n period’s observations • Example: for n = 3: Sales for Jan through March were 100, 110, 150 • April forecast = (100+110+150)/3 = 120
Evaluating Forecasts • Concept: Forecast worth function of how close forecasts are to observations • Mean Absolute Deviation (MAD) • MAD = sum of absolute value of forecast errors / number of forecasts (e.g. periods) • MAD is the average of the absolute value of all of the forecast errors.
Weighted Moving Averages • This period’s forecast = Weighted average of past n period’s observations • Example: for n = 3: Sales for Jan through March were 100, 110, 150 • Suppose weights for last 3 periods are: .5 (March), .3 (Feb), and .2 (Jan) • April forecast =.5*150+.3*110+.2*100 = 128
Exponential Smoothing • New Forecast = Last period’s forecast + alpha * (Last period’s actual observation - last period’s forecast) • Mathematically: F(t) = F(t-1) + alpha * [A(t-1) - F(t-1)], where F is the forecast; A is the actual observation, and alpha is the smoothing constant -- between 0 and 1 • Example: F(t-1) = 100; A(t-1) = 110; alpha = 0.4 -- Find F(t) • F(t) = 104 • Can add parameters for trends and seasonality
Trend Projections • Use Linear regression • Model: yhat = a + b* x • a = y-intercept: forecast at period 0 • b = slope: rate of change in y for each period x • Example: Sales = 100 + 10 * t, where t is period • For t = 15, Find yhat -- • yhat = 250 • Can find and a and b via Method of Least Squares
Linear Regression • Model: yhat = a + b1 * x1 + b2 * x2 + … + bk * xk • a = y-intercept • bi = slope: rate of change in y for each increase in xi, given that other xj’s are held constant • Example: College GPA = 0.2 + 0.5 HS GPA + 0.001 HS SAT • For a HS student with a 3.0 GPA and 1200 SAT - what is the forecast? • The forecast college GPA = 2.90 • Can find a, b1, and b2 via Method of Least Squares