Beruflich Dokumente
Kultur Dokumente
Pard Teekasap
Southern New Hampshire University
Outline
1. Definition of the Delays
2. Material Delays
3. Information Delays
4. Variable Delay Time
5. Estimate the Delays
Pre-class Question
• Consider the market for agricultural
commodities such as pork. What is the
average delay between a rise in the price of
pork and the resulting increase in pork
supply?
• If there is an unanticipated increase in the
inflation rate, how long will it take for the
forecasts of the experts to adjust to the new
rate?
Define the Delays
• A delay is a process whose output lags behind its
input in some fashion
• Since the input differs from output, there must
be a stock inside the process to accumulate the
differences
• Material delay = a delay process of the physical
flow of material
• Information delay = the gradual adjustment of
perception or beliefs
Delays have stocks
Material Delays
• In many cases, the outflows are constrained by
limited resources. These resources must be
modeled explicitly
• The important aspects of delays
– What is the average length of the delay?
– What is the distribution of the output around the
average delay time?
What is the average length of the delay?
150
D
100
B
50
C
0
0 1 2 3
Time (multiples of average delay time)
A = Pipeline delay; B = First-order delay;
C = Third-order delay; D = Twelfth-order delay
Pipeline Delay
• Also known as transportation lag
• The delay time is constant
• The order of exit from the delay is precisely
the same as the order of entry
• E.g. Auto assembly line
• Outflow(t) = Inflow(t-D)
Structure of Pipeline Delay
First-Order Material Delay
• There is mixing and variation in the individual
processing times, causing some variance in the
distribution of delivery
• Perfect mixing = the order of entry is ignored
• Outflow = Material in Transit/D
Structure of First-order Material Delay
Behavior of First-order Material Delay
Delay Inflow and Outflow
100
% of Unit Pulse/Time Period
75
50
Inflow Outflow
25
0
0 1 2 3
Time (multiples of average delay time)
Stock of Material in Transit
100
(% of pulse quantity)
75
Material in Transit
Material in Transit
50
25
0
0 1 2 3
Time (multiples of average delay time)
Higher-order Material Delay
• The delay consists of multiple stages of
processing in which items flow sequentially
from one stage to the next, but where each
stage introduces some mixing
• The same as cascading first-order delays
together
Structure of Higher-order Delay
Nth-order Material Delay
• A delay with n stages, each with 1/n of the
total delay time
• The higher the order of the delay, the less
mixing and the smaller the variance of the
output
• An infinite-order delay preserves the order of
entry and permits no mixing = pipeline delay
Pulse response of a 3rd-order delay
Delay Inflow and Outflow
100
% of Unit Pulse/Time Period
75
50
Inflow Outflow
25
0
0 1 2 3
Time (multiples of average delay time) Stock of Material in Transit
100
(% of pulse quantity)
75
Material in Transit
50
25
0
0 1 2 3
Time (multiples of average delay time)
Pulse response of 3rd-order delay by stage
Delay Inflow and Outflow
300
Stage 1
Exit Rate
200
Stage 2
Inflow
Exit Rate
100
Outflow
0
0 1 2 3
Time (multiples of average delay time) Stock of Material in Transit
100
Stage 1
50
Stage 2
25
Stage 3
0
0 1 2 3
Time (multiples of average delay time)
Little’s Law
• How big will the stock in transit be for any
given delay and inflow, suppose the inflow has
been constant long enough for the delay to
reach equilibrium?
• The equilibrium stock in transit for a delay is
always DI units, regardless of the probability
distribution of the outflow
Information Delays
• Why do perceptions and forecasts involve
delays?
• It takes time to gather the information needed
to form judgments
• People don’t change their mind immediately
on the receipt of new information
• Need time to adjust emotionally to a new
situation
Adaptive expectation
• Also known as Exponential smoothing
• Simplest information delay and one of the
most widely used
• The belief gradually adjusts to the actual value
of the variable
• It’s also known as a first-order information
delay or first-order exponential smoothing
Structure of adaptive expectation
Response of adaptive expectations to a step
change
200
Perceived and Actual Values
Input:
Actual Value
Output:
Perceived Value
(units)
150
100
0 1 2 3
Time (multiples of average delay time)
100
75
Units/Time Period
50
Change in Perceived Value
25
0
0 1 2 3
Time (multiples of average delay time)
Adaptive expectation eliminate short-term
noise
1500
Order Rate
1250
Units/Day
1000
750 Expected
Order Rate
500
0 50 100 150 200 250 300
Days
Higher-order information delays
• In a first-order information delay, the output
responds immediately to a change in the input
• In many cases, beliefs begin to respond after
some time has passed
• The delay between the actual state and the
decisions involves multiple stages
• Similar to the higher-order material delay, we
cascade first-order smoothing structures
Structure of the 3rd-order info delay
Response of higher-order delays to a step
input
200
Perceived and Actual Values
Input:
Actual Value Output:
Perceived Value
1st
(units)
Order
150
3rd
Order
12th
Order
100
0 1 2 3
Time (multiples of average delay time)
Variable Delay Time
• In many cases, the delay time is not fixed. It can
varies both exogenously and endogenously
• If you’re the only one at the ATM, you can get a
cash by about a minute
• However, the delay time to get cash increases if
there are people ahead of you
• The rate at which people joint the line also
depends on how many people are in line
Ratchet Effects
• People moves with positive things faster than
negative things
• Aggregate consumption expenditures rise
faster than they fell as income fluctuated over
the business cycle
• People perceive job insecurity quickly and
have a long memory of that
Nonlinear time constant in job insecurity
Memory of
Layoffs
Change in
Layoff
Memory
0.10
(fraction of workforce/week)
+ - -
Layoff
Rate Memory
Adjustment
Layoff Rate
Time
- + + +
0.05 Memory of
Layoffs DI DD
Layoff
Rate
0.00
0 20 40 60 80 100
Weeks
Estimating delays from data
• Econometric techniques can be used for
estimating lags from time series include the
Koyck or geometric lag, polynomial distributed
lags, rational distributed lags. And ARIMA
model
• You need to trade off the flexibility of the
formulation against the number of parameters
to be estimated in choosing the method
Don’t put the regression equation in the
model
• You should replace the estimated distributed
lag with the material or information delay that
best matches the estimated lag because
– Econometric techniques are for discrete time, but
system dynamics are for continuous time
– The delay time may be incorporated as an
endogenous variable
– Regression equations do not distinguish between
material and information delays
Estimation the lagged response of natural gas
supply (3rd-order)
Construction lag for capital plant (2nd-order)
5
2nd Order Delay
4
% of Unit Pulse/Month
Survey
3 Data
1
Mean Delay
(16.7 months)
0
0 12 24 36 48
Month
Estimating delays when no available data