Transcript slides_cycles1_09

The Global Economy
Business Cycle Indicators
© NYU Stern School of Business
Bloomberg calendar
• Why do we look at this stuff?
http://www.bloomberg.com/markets/ecalendar/index.html
Plan of attack
• Bloomberg calendar 
• Business cycle overview
• Pictures
• Forecasting
• Good indicators (the “cross-correlation function”)
• Forecasting revisited
• What have we learned?
Business cycle overview
Current Conditions
Indicators
Statistical
Analysis
Theory:
AS & AD
Future Conditions
Monetary Policy
US GDP
Source: BEA via FRED.
US GDP growth (year-on-year)
Source: BEA via FRED.
Components of US GDP growth
Source: BEA via FRED.
Components of US GDP growth
Source: BEA via FRED.
-1000
-500
Thousands
0
500
Employment (month-to-month change)
1990
Source: BLS.
1992
1994
1996
1998
2000
2002
2004
2006
2008
1000
0
500
Thousands
1500
2000
Housing starts
1990
Source: Census.
1992
1994
1996
1998
2000
2002
2004
2006
2008
Recession futures
• Probability recession in 2009 (2 qtrs negative growth)
Source: Intrade.com.
Forecasting
• Can we forecast future economic conditions?
– Use information about the past & present to predict the future
• Basic idea: use patterns in the data
– In the past, “this” was followed by “that”
Forecasting: example
• Example: growth in industrial production (“IP”)
γt,t+k = log(IPt+k/IPt)/k = k-period growth rate in IP
• Why? How?
Forecasting: regressions
• Reminder: we’re forecasting
γt,t+k = log(IPt+k/IPt)/k = k-period growth rate in IP
[set k=12 months?]
• Estimate the regression
γt,t+k = a + b xt + residual
xt is an “indicator” of your choice
• Standard software produces estimates of a and b
• Note timing!!
Forecasting: fancy regressions
• Basic regression
γt,t+k = a + b xt + residual
– γt,t+k is what we’re trying to forecast
– xt is an “indicator” of your choice
• Variations
– Use more than one x
– Add lags of x
– Add lags of γ
– Whatever works!
Forecasting: forecasts
• Recall: we’ve estimated a and b in
γt,t+k = a + b xt + residual
• Calculate predicted future growth rate
γt,t+k = a + b xt
– xt = value now (t) of indicator x
– γt,t+k = predicted growth from now (t) to the future (t+k)
• Make sure you understand the timing
Identifying good indicators
• What would you suggest? Why?
Identifying good indicators
• How do you find good indicators?
– Forecasting requires indicators that lead what you’re forecasting
– Ask friends, read reports, look at “cross-correlation function”
Identifying good indicators
Lags IP
-1.00
-1.00
-0.50
-0.50
0.00
0.00
0.50
0.50
Leads IP
1.00
1.00
Correlations
for Random
Indicator X
Nonfarm
Employment
-20
-10
0
10
Lag in Months Relative to Industrial Production
20
Identifying good indicators
• Cross-correlation function (“ccf”)
– Correlations between two variables at different times
ccf(k) = Corr(xt,yt-k)
[plot this against k]
– If k<0: x leads y [or y lags x]
– If k>0: x lags y [or y leads x]
Identifying good indicators
• Pictures: plot ccf(k) v k
– y = IP growth
– x = indicator
– Does indicator lead or lag IP growth?
Identifying good indicators
• Mechanics of ccf calculations: see spreadsheet
Does employment lead or lag?
1.00
Leads IP
Lags IP
0.50
What is
this dot?
-1.00
-1.00
-0.50
-0.50
0.00
0.00
0.50
1.00
Nonfarm Employment
-20
-10
0
10
Lag in Months Relative to Industrial Production
20
Computing cross-correlations
Date
x(t)
y(t)
1
2.43
8.47
2
1.19
2.29
3
0.13
7.36
4
0.56
6.39
5
0.38
6.02
6
0.96
0.22
7
1.87
3.60
Reminder:
• ccf(k) = corr[x(t),y(t-k)]
For k = 0:
• ccf(0) = corr[x(t),y(t)]
Use data marked
• Red for x
• Blue for y
Computing cross-correlations
Date
x(t)
y(t)
1
2.43
8.47
2
1.19
2.29
3
0.13
7.36
4
0.56
6.39
5
0.38
6.02
6
0.96
0.22
7
1.87
3.60
Reminder:
• ccf(k) = corr[x(t),y(t-k)]
For k = +1:
• ccf(1) = corr[x(t),y(t-1)]
• Means: x lags y
Use data marked
• Red for x
• Blue for y
Computing cross-correlations
Date
x(t)
y(t)
1
2.43
8.47
2
1.19
2.29
3
0.13
7.36
4
0.56
6.39
5
0.38
6.02
6
0.96
0.22
7
1.87
3.60
Reminder:
• ccf(k) = corr[x(t),y(t-k)]
For k = –1:
• ccf(-1) = corr[x(t),y(t+1)]
• Means: y lags x
Use data marked
• Red for x
• Blue for y
Does employment lead or lag?
Lags IP
-1.00
-1.00
-0.50
-0.50
0.00
0.00
0.50
0.50
Leads IP
1.00
1.00
Nonfarm Employment
-20
-10
0
10
Lag in Months Relative to Industrial Production
20
Unemployment?
Lags IP
-10
10
1.00
Leads IP
-1.00
-1.00
-0.50
-0.50
0.00
0.00
0.50
0.50
1.00
Unemployment Rate
-20
0
Lag in Months Relative to IP
20
New claims for un ins?
Lags IP
-10
10
1.00
Leads IP
-1.00
-1.00
-0.50
-0.50
0.00
0.00
0.50
0.50
1.00
Unemployment: New Claims
-20
0
Lag in Months Relative to IP
20
Housing starts?
Lags IP
-1.00
-1.00
-0.50
-0.50
0.00
0.00
0.50
0.50
Leads IP
1.00
1.00
Housing Starts
-20
10
0
-10
Lag in Months Relative to Industrial Production
20
Building permits?
Lags IP
-1.00
-1.00
-0.50
-0.50
0.00
0.00
0.50
0.50
Leads IP
1.00
1.00
Building Permits
-20
-10
0
10
Lag in Months Relative to Industrial Production
20
Consumer sentiment?
Lags IP
-1.00
-1.00
-0.50
-0.50
0.00
0.00
0.50
0.50
Leads IP
1.00
1.00
Consumer Sentiment
-20
-10
0
10
Lag in Months Relative to Industrial Production
20
S&P 500 index?
Lags IP
-1.00
-1.00
-0.50
-0.50
0.00
0.00
0.50
0.50
Leads IP
1.00
1.00
S&P 500 Index
-20
-10
0
10
Lag in Months Relative to Industrial Production
20
Yield spread?
1.00
1.00
Yield Spread (10y - Fed Funds)
Lags IP
-1.00
-1.00
-0.50
-0.50
0.00
0.00
0.50
0.50
Leads IP
-20
-10
0
Lag Relative to IP
10
20
Forecasting flow chart
1.
Identify good indicators
–
2.
Transform them if appropriate
–
3.
4.
Use cross-correlation function, ask friends, whatever works
Do you use the level? Growth rate? Change?
Put them in a regression
–
Tells you relation between indicator and variable being
forecast
–
How long a sample do you use? [1985]
Use the regression coefficients and current value of
indicator to construct forecast
How well do we do?
• Forecasts have content
– Typical R2 > 0, < 0.50
– Most of what happens is not predicted
• Therefore: have a contingency plan
What have we learned?
Takeaways
• Good indicators tell us something about the future
• Even the best forecasts leave lots of uncertainty
• Useful tools:
– Regressions
– Cross-correlation function
Enjoy the break
• See you in a couple weeks