Вы находитесь на странице: 1из 26

ECON 332

Business Forecasting Methods


Time Series Decomposition Models
Chapter 6
Prof. Kirti K. Katkar
6-2

Time Series Decomposition Models

• One of the oldest and most popular models


– In many situations they provide excellent forecasts
– They are easy to understand and explain
– They provide concrete measurements for factors that otherwise cannot
be quantified individually and in an interactive manner
• Many business and economic phenomena can be modeled with
time series decomposition methods since they all contain
– Long-term trend
– Seasonal fluctuations
– Cyclical movements
– Random fluctuations
• All business cycles can be modeled and explained by these
models
6-3
Overview of Business Cycles
•Business cycles are long-term wavelike fluctuations in the level of
business and economic activity

•Activity level from A to B is expansion phase or Growth


•Activity level from B to C is contraction phase or Recession
6-4

Business Cycle Indicators

• Index of leading economic indicators


• Index of coincident economic indicators
• Index of lagging economic indicators
6-5

U. S. Business Cycle Indicators


6-6
U.S. Leading Index

Official Business Cycles in the U.S. – National Bureau of Economic Research (NBER)
P – Peak, end of expansion, beginning of recession
T – Trough, end of recession, beginning of expansion
- Indicates leads (-) and lags (+) in months
6-7
U.S. Coincident Index

Official Business Cycles in the U.S. – National Bureau of Economic Research (NBER)
P – Peak, end of expansion, beginning of recession
T – Trough, end of recession, beginning of expansion
- Indicates leads (-) and lags (+) in months
6-8
U.S. Lagging Index

Official Business Cycles in the U.S. – National Bureau of Economic Research (NBER)
P – Peak, end of expansion, beginning of recession
T – Trough, end of recession, beginning of expansion
- Indicates leads (-) and lags (+) in months
6-9

Basic Time Series Decomposition Models

Let T – Long-term trend


S – Seasonal Adjustment Factor
C – Cyclical Adjustment Factor
I – Irregular or Random Factor
and Y – Variable to be forecast; then
• The Basic time series decomposition (multiplicative) model
Y=TxSxCxI
• To build the model and forecast we need to decompose the
given time series Yi into its components - T, S, C and I
6-10 Forecasting with Basic Time Series
Decomposition Models
• First we estimate the seasonal indices (SI)
SI = Actual Value Y/ Deseasonalized Value of Y (CMA)
where CMA is the corresponding centered moving average of the
actual value
• The long term trend is the trend in deseasonalized data. It is
also called centered moving average trend (CMAT) since the
deseasonalized data is the CMA of the original data
• The cyclical component cycle factor CF = CMA/ CMAT
represents the gradual wavelike movement around the trend
line
• I the irregular component is assumed to be = 1 in most cases
• The forecast then is
FY = (CMAT)(SI)(CF)(I)
6-11 Private Housing Starts (PHS) Time Series
(Unit 000)
6-12

Estimating CMAs and SIs


6-13

Deseasonalizing Data and Determining


Seasonal Indices
• Moving average(MA) removes short term fluctuations – both
seasonal and random
• For deseasonalizing the MAs should contain same # of periods
as the seasonality that is evident – 4 for quarterly seasonality,
12 for monthly seasonality
• For quarterly data
MAt = (Yt-2 + Yt-1 + Yt +Yt+1)/4
• For monthly data
MAt = (Yt-6 + Yt-5 + Yt-4 + ….. + Yt + Yt+1 + … + Yt+5)/12
• Centered moving average (CMA)
CMAt = (MAt + MAt+1)/2
• Seasonal factor (SF)
SFt = Yt/ CMAt
• Seasonal Index (SI) is the normalized mean of corresponding
SFs seasonal factors
6-14 Illustration: Deseasonalizing Quarterly Data and
Determining Seasonal Indices
Year/ Time Index X MA CMA SF
Quarter
1Q1 1 10 Missing Missing Missing
1Q2 2 18 Missing Missing Missing
1Q3 3 20 15 15.25 1.31
1Q4 4 12 15.5 15.75 0.76
2Q1 5 12 16 Missing Missing
2Q2 6 20 Missing Missing Missing

MA3 = (10+18+20+12)/4 = 60/4 = 15 CMA3 = (15 +15.5)/2 = 30.5/2 = 15.25


MA4 = (18+20+12+12)/4 = 62/4 = 15.5 CMA4 = (15.5 + 16)/2 = 31.5/2 = 15.75
MA5 = (20+12+12+20)/4 = 64/4 = 16 SF3 = 20/15.25 = 1.31 and SF4 = 12/15.75 = 0.76
6-15 PHS - Original and Deseasonalized Time
Series
6-16 Decomposition of Private Housing Starts (PHS) Time Series
6-17

Estimating the Long-term


Trend
6-18

The Long-term Trend


• The long-term trend is estimated from the deseasonalized data.
CMA is the only series that remains after the seasonality and
the irregular fluctuations have been removed.
• Thus CMA = f (Time) = a + b (Time),
where Time = 1 for the first period in the time series and it
increases by 1 for each subsequent period – month or quarter.
• The values of a and b are estimated using the computerized
linear regression programs
• Once the trend equation has been determined, it is used to
generate the trend value of the CMA for the historical and the
forecast periods. This yields a new series – CMAT: centered
moving average trend
6-19 PHS Series Illustration with CMA and CMAT

In this PHS case the equation for the trend line


is PHSCMAT = 237.51 + 0.313 (TIME), a slightly
positive trend indicated by the +ve slope ) 0.313
6-20 Decomposition of Private Housing Starts (PHS) Time Series
6-21

Estimating the Cyclical


Factor (CF)
6-22
Estimating the CFs
• CF is the wave-like movement around the long-term trend
• CF = CMA/ CMAT
• CF > 1 indicates that the deseasonalized value for the period is
above the long-term of the data
• CF < 1 indicates that the deseasonalized value for the period
is below the long-term of the data
• CFs are very useful in analyzing the series and represent the
value add of the time series decomposition models
• The length and amplitude of previous cycles should give a
clue on the next turning points – both in terms of timing and
magnitude
6-23
CF for PHS
6-24 Decomposition of Private Housing Starts (PHS) Time Series
6-25

Figure 6-7
6-26
PHS Actuals and Forecasts

Вам также может понравиться