Forecasting Models

1
Forecasting Models

• Static vs. Dynamic

2
Basics

• Same underlying causal system that existed in
  past will continue.
• Aggregate Forecast are better than disaggregate
  ones.
• Forecast accuracy decreases as time horizon
  decreases.

Approaches to Forecasting
Judgmental Forecasts Subjective inputs from
various sources Consumer surveys managers
sales staff (Delphi Survey) Expert opinion-
Fortune Tellers
Time-Series Forecasts Ordered sequence of
observations taken at regular intervals over a
period of time. Goal Identify the behavior of
the series and project this behavior to
future. Average Trend Seasonality Cycles-
Random Fluctuations
3
Visual Investigation
RANDOM FLUCTUATIONS AROUND AN AVERAGE
TREND
time
time
SEASONALITY
time
Short term
4
Static Model , Simple Average

• Basic Assumption No change in demand, all
  fluctuations are random.
• Forecast Average
• A measure of match between forecast and actual
  data is MAD
• Mean Absolute Deviation
• Average (absolute value of (sales-forecast) for
  the overall data range)

5
Static Model Modeling Trend

• Assumption There is a linear trend in data.
• Plot the data and add trend line (linear).
• Intercept
• Slope Trend
• First Period Forecast Intercept Trend
• Forecast Previous period forecast Trend

6
Static Model with SeasonalitySeasonality(Linear
Trend)
Seasonality Index
3 yr Average / 3 yr. Average of for Jan
monthly sales
Forecast Seasonality index (Previous Period
Forecast Trend)
3 yr average of monthly sales
7
Dynamic Model Exponential Smoothing
Suppose we have just observed actual sales for
January, and we want to forecast February. Hence,
at hand we have our forecast for January (which
we fixed before we observed sales for period t)
and our observation for period January
Forecast for (tFebruary) (1-a) (Forecast for
January) a (Actual in January) Forecast
for January a(Actual January - Forecast for
January )
Forecast Error
Smoothing Constant (a) is determined by
judgments Trade-off between responding to random
fluctuations vs. identifying real changes.
Popular technique since it requires minimum data
storage.
8
Exponential Smoothing with Trend

• First Period Forecast
• Intercept Slope (from trendline)
• Next Period Forecast
• Average (1-a)Previous Forecast aPrevious
  Sales
• Slope b D in Averages (1-b)Previous Trend
• Forecast Average Trend
• To incorporate seasonality multiply forecast with
  seasonality factor.