The Optimum Pokemon Portfolio and Principal Component Decomposition (PCD) using R

I have very recently completed the Stanford Lagunita online course on Statistical Learning, and Tibrishani & Hastie have taught me a great deal about Principal Components.  No learning is complete without exercises, however, so I have found a wonderful data set that seems popular, the attacks and weaknesses of Pokemon.  (I am, admittedly, not a pokemon player, so I have had to ask others to help me understand some of the intricacies of the game.)

Principal Component Decomposition:

First and foremost, principal component decomposition finds the direction that maximizes variation in the data.  At the same time, this can be said to be the eigenvalue of the data, the direction which best describes the direction of the data.
For example, if there is a spill of dirt on a white tile floor, the direction of the spill (eigenvalue) would always be the direction the dirt is most widely spread (principal component).

After looking at the beautiful charts used in the link above, I realized this would be very interesting to do a PCD on. What Pokemon are most similar and which are most different in terms of strengths and weaknesses? To find out we will break it into its principal components, and find out in which directions the data is spread out.

Pokemon can vary along 18 dimensions of strengths and weaknesses, since there are 18 types of Pokemon. This means there can be up to 18 principal components. We are not sure which principal components are useful without investigation. We show below how much variation is explained by each type of Pokemon. There doesn’t appear to be any clear point where there the principal components drop off in their usefulness, perhaps the first 3 or the first 5 seem to capture the most variation.  The amount of variation captured by each principal component is outlined below.

Let us now look at the principal components of the Pokemon attack/weakness chart directly.  We can visualize them in a biplot, where the arrows show the general attacking direction of the pokemon and the black labels show the defending labels.  The distance from the center of biplot shows the deviation of that pokemon type from the central eigenvalue/principal component.  Labels that are close together are more similar than those further apart.

So for example, Ghost attacks (arrows) are closely aligned with Ghost defence (black label) and Dark defence (black label).  In general, the Pokemon that are most different in defence is Fighting and Ghost, and still again distinct from Flying and Ground defence.  This suggests that if you wanted a Pokemon portfolio that would be very resilient to attack, you would want Fighting/Ghost types.  If you want a variety of attacks, you might want to look into Ghost/Normal types or Grass/Electric.

Keep in mind together these only explain about 35.5% of the variation of Pokemon types, there are other dimensions in which Pokemon vary.  I expected fire and water to be more clearly different (and they are very distinct, they go opposite directions for a long distance from the center!), but they are less distinct than ghost/normal.

The Optimum Pokemon Portfolio:

This lead me to wonder what type of pokemon portfolio would be best against the world, something outside the scope of the Statistical Learning course but well within my reach as an economist.  Since I don’t know what the pokemon-world looks like, I assumed the pokemon that show up are of a randomly and evenly selected type. (This is a relatively strong assumption, it is likely the pokemon encounters are not evenly distributed among the types).  The question is then, what type of pokemon should we collect to be the best against a random encounter, assuming we simply reach into our bag and grab the first pokemon we see to fight with?

First, I converted the matrix of strengths and weaknesses above into one that describes the spread of the strength-weakness gap, that is to say, if Water attacks Fire at 200% effectiveness, and defends at 50% effectiveness, a fight between the Water and Fire is +150% more effective than a regular pokemon attack (say Normal to Normal or Ice to Ice). Any bonuses a pokemon may have against its own type was discarded, because it would be pointless.  The chart for this, much like the wonderful link that got me the data in the first place, is here, where red is bad and blue is good:

Then I added the strength-weakness gap together for each type of pokemon, which assumes that the pokemon are facing an a opponent of a random type.  According to this then, the most effective type of pokemon are on average:

Type                              Effectiveness
Steel                               0.22222222
Fire                                0.11111111
Ground                              0.11111111
Fairy                               0.11111111
Water                               0.08333333
Ghost                               0.08333333
Flying                              0.05555556
Electric                            0.00000000
Fighting                            0.00000000
Poison                             -0.02777778
Rock                               -0.02777778
Dark                               -0.02777778
Ice                                -0.08333333
Dragon                             -0.08333333
Normal                             -0.11111111
Psychic                            -0.11111111
Bug                                -0.11111111
Grass                              -0.19444444

That is to say, Steel pokemon, against a random opponent, will on average be 22% more effective.  (This is the mean, not the median.) And against a random opponent a Grass pokemon will be expected to be 19% less effective than a Fighting pokemon, shockingly low. Amusingly, Normal pokemon are worse than normal (0) against the average pokemon.

This does not mean you ONLY want Steel pokemon because you could come up with an opponent that is strong against Steel. Nor do you want to entirely avoid Grass pokemon, since they are very strong against many things that Steel is weak against. Merely that if you’re willing to roll the dice, a Steel pokemon will probably be your best bet.  Trainers do not want to take strong risks, trainers are risk averse.  You want to maximize your poke-payoff while minimizing how frequently you face negatively stacked fights. The equation for this is:

$Maximize: \ \mu * vars - \delta * t(vars) * cov * vars + \lambda*(1- t(ones) * vars) \ wrt. \ vars$

Where $\mu$ is your vector of payoffs in the table above, $\delta$ is your risk aversion, cov is the covariance matrix of the differenced pokemon data set, and vars is your portfolio selection which must add up to one hundred percent.

How risk averse are you?  You could be very risk averse and want to never come across a bad pokemon to fight, or you could love rolling the dice and only want one type of pokemon. So I have plotted the optimal portfolio for many levels of risk-tolerance.  It is a little cluttered, so I have labelled them directly as well as in the legend.

The visualization is indeed a little messy, but as you become more risk averse, you add more Electric, Normal, Fire, and Ice pokemon (and more!) to help reduce the chance of a bad engagement.  In order to do this, one reduces the weight we put on Steel, Ground, and Fairy pokemon, but doesn’t eliminate them entirely.  Almost nothing adds Dragon, Ghost, Rock. or Bug pokemon, they are nearly completely dominated by other combinations of pokemon types.

I’ve plotted two interesting portfolios along the spectrum of risk aversion below. They include one with nearly no risk aversion (0.001), and one with high risk aversion (10).

This slideshow requires JavaScript.

Of course, most importantly of all, regardless of your Pokemon and your interest in being “the very best”, you should still pick the coolest Pokemon and play for fun.

Code is included below:

#Data from: https://github.com/zonination/pokemon-chart/blob/master/chart.csv
#write.csv(chart, file="/home/bsweber/Documents/poke_chart.csv")
poke_chart<-poke_chart[,-1]
# library(devtools)
# install_github("vqv/ggbiplot", force=TRUE)
library(ggbiplot)
library(reshape2)
library(ggplot2)
library(ggrepel)

poke_chart<-as.matrix(poke_chart)
differences <- (poke_chart-1) - (t(poke_chart)-1)
diag(differences)<-0
rownames(differences)<-colnames(differences)
core <- poke_chart
rownames(core)<-colnames(poke_chart)

poke_pcd<-prcomp(core, center=TRUE, scale=TRUE)
plot(poke_pcd, type="l", main="Pokemon PCD")
summary(poke_pcd)
biplot(poke_pcd)

poke_palette<-c("#A8A878", "#EE8130", "#6390F0", "#F7D02C", "#7AC74C", "#96D9D6", "#C22E28", "#A33EA1", "#E2BF65", "#A98FF3", "#F95587", "#A6B91A", "#B6A136", "#735797", "#6F35FC", "#705746", "#B7B7CE", "#D685AD")

ggbiplot(poke_pcd, labels= rownames(core), ellipse = TRUE, circle = TRUE, obs.scale = 1, var.scale = 1) +
scale_color_discrete(name = '') +
theme(legend.direction = 'horizontal', legend.position = 'top')
#Score plot is for rows, attack data. loading lot is for columns, defense data.  So bug and fairy have similar attacks (shown by rays), similar defences (shown by points). Ghost and normal have almost identical defences, but different attacks.
ggbiplot(poke_pcd, labels= colnames(core), ellipse = TRUE, circle = TRUE, obs.scale = 1, var.scale = 1, choice=c(2,3)) +
scale_color_discrete(name = '') +
theme(legend.direction = 'horizontal', legend.position = 'top')  #Score plot is for rows, attack data. loading lot is for columns, defense data.
ggbiplot(poke_pcd, labels= colnames(core), ellipse = TRUE, circle = TRUE, obs.scale = 1, var.scale = 1, choice=c(5,6)) +
scale_color_discrete(name = '') +
theme(legend.direction = 'horizontal', legend.position = 'top')  #Score plot is for rows, attack data. loading lot is for columns, defense data.
ggbiplot(poke_pcd, labels= colnames(core), ellipse = TRUE, circle = TRUE, obs.scale = 1, var.scale = 1, choice=c(7,8)) +
scale_color_discrete(name = '') +
theme(legend.direction = 'horizontal', legend.position = 'top')  #Score plot is for rows, attack data. loading lot is for columns, defense data.

cov_core<- t(differences-mean(differences)) %*% (differences-mean(differences)) #Make the Cov. Matrix of differences.
cov_core[order(diag(cov_core), decreasing=TRUE),order(diag(cov_core), decreasing=TRUE)]
ones<-as.matrix(rep(1,18))
vars<-as.matrix(rep(1/18, times=18))
mu<-t(as.matrix(apply(differences/18, 1, sum))) #Average rate of return over 18 pokemon types.

data.frame(mu[,order(t(mu), decreasing=TRUE)]) #Table of Pokemon Types

colnames(mu)<-colnames(core)
delta<- 1  #risk aversion parameter

out<- matrix(0, nrow=0, ncol=18)
colnames(out)<-colnames(core)
for(j in 1:1000){
delta<-j/100
Dmat <- cov_core * 2 * delta
dvec <- mu
Amat <- cbind(1, diag(18))
bvec <- c(1, rep(0, 18) )
qp <- solve.QP(Dmat, dvec, Amat, bvec, meq=1)
pos_answers<-qp$solution names(pos_answers)<-colnames(poke_chart) out<-rbind(out, round(pos_answers, digits=3)) } df <- data.frame(x=1:nrow(out)) df.melted <- melt(out) colnames(df.melted)<-c("Risk_Aversion", "Pokemon_Type", "Amount_Used") df.melted$Risk_Aversion<-df.melted$Risk_Aversion/100 qplot(Risk_Aversion, Amount_Used, data=df.melted, color=Pokemon_Type, geom="path", main="Pokemon % By Risk Aversion") + # ylim(0, 0.175) + scale_color_manual(values = poke_palette) + # geom_smooth(se=FALSE) + geom_text_repel(data=df.melted[df.melted$Risk_Aversion==8.5,], aes(label=Pokemon_Type, size=9, fontface = 'bold'), nudge_y = 0.005, show.legend = FALSE)

# Another plot that is less appealing
# matplot(out, type = "l", lty = 1, lwd = 2, col=poke_palatte)
# legend( 'center' , legend = colnames(core), cex=0.8,  pch=19, col=poke_palatte)
pie(tail(out, 1), labels= colnames(out), col=poke_palette)

df_1<-data.frame(matrix(out[1,], ncol=1))
colnames(df_1)<-c("Percentage")
df_1$Pokemon_Type<-colnames(out) ggplot(data=df_1, aes(x=Pokemon_Type, y=Percentage, fill=Pokemon_Type))+ geom_bar(stat="identity", position=position_dodge()) + scale_fill_manual(values = poke_palette)+ ggtitle("Pokemon Portfolio With Almost No Risk Aversion") df_2<-data.frame(t(tail(out,1))) colnames(df_2)<-c("Percentage") df_2$Pokemon_Type<-colnames(out)

ggplot(data=df_2, aes(x=Pokemon_Type, y=Percentage, fill=Pokemon_Type))+
geom_bar(stat="identity", position=position_dodge()) +
scale_fill_manual(values = poke_palette) +
ggtitle("Pokemon Portfolio With Very Strong Risk Aversion")

cov_core[order(diag(cov_core), decreasing=TRUE),order(diag(cov_core), decreasing=TRUE)]

melt_diff<-melt(t(differences))
melt_diff$value<- factor(melt_diff$value)
N<-nlevels(melt_diff$value) simplepalette<-colorRampPalette(c("red", "grey", "darkgreen")) ggplot(data = melt_diff, aes(x=Var1, y=Var2, fill=value) ) + geom_tile()+ scale_fill_manual(values=simplepalette(9), breaks=levels(melt_diff$value)[seq(1, N, by=1)], name="Net Advantage" )+
xlab("Opponent") +
ylab("Pokemon of Choice")
Statistics

Multiple Linear Regression in R

This slideshow requires JavaScript.

In the previous exercise: Why do we need N-2?, I show a simple 1 dimensional regression by hand, which is followed by an examination of sample standard errors.  Below I make more extensive use of R (and an additional package) to plot what linear regression looks like in multiple dimensions. This generates the images above, (along with several others).  This illustrates that linear regression remains flat even in N dimensions, the surface of the regression is linear in coefficients.

As a class exercise, I ask that you consider different pairs dependent variables that are  functions of one another. What happens if the function is linear? What happens if the function is nonlinear, for example, $cos(x_1)=x_2$? Examine what happens to the surface of your regression as compared to the shape of the relationship you are investigating.  Is there a way you can contort the regression estimate into a curved surface to better match?  Why or why not?

install.packages(“plot3D”) # we need 3d plotting
library(“plot3D”, lib.loc=”~/R/win-library/3.1″) #Load it into R’s current library, may vary by computer.

set.seed(2343) #ensures replicatation. Sets seed of random number generators.
n<-25 #number of samples
x_1<-rnorm(n) #Our x’s come from a random sampling of X’s.
x_2<-rnorm(n)
b_0<-10
b_1<-3 #Those cursed jello puddings are associated with increased crime. Linear regression is supportive of association- not causation.
b_2<-(-3) # But student transit programs are associated with a decline in crime.
u<-rnorm(n)
y<-b_0+b_1*x_1+b_2*x_2+u #This is defining our true Y. The true relationship is linear.

#look at data in each dimension
plot(x_1,y)
plot(x_2,y)
#look at data overall
points3D(x_1,x_2,y,xlab=”x_1″,ylab=”x_2″,zlab=”y”,phi=5) #look at data. phi/theta is tilt.

fit<-lm(y~x_1+x_2)  #fit it with a linear model, regressing y on x_1, x_2

#Make a surface
x_1.pred <- seq(min(x_1), max(x_1), length.out = n)
x_2.pred <- seq(min(x_2), max(x_2), length.out = n)
xy <- expand.grid(x_1=x_1.pred, x_2=x_2.pred)
y.pred <- matrix (nrow = n, ncol = n, data = predict(fit, newdata = data.frame(xy), interval = “prediction”))

summary(fit) #view output of variables.

fitpoints<-predict(fit)  #get predicted points, needed to make a surface.

scatter3D(x_1,x_2,y,xlab=”x_1″,ylab=”x_2″,zlab=”y”,phi=5 , surf=list(x = x_1.pred, y = x_2.pred, z = y.pred, facets = NA, fit = fitpoints)) #look at data. phi/theta is tilt.
scatter3D(x_1,x_2,y,xlab=”x_1″,ylab=”x_2″,zlab=”y”,phi=45, surf=list(x = x_1.pred, y = x_2.pred, z = y.pred, facets = NA, fit = fitpoints)) #From straight on it is a flat plane, residuals are highlighted
scatter3D(x_1,x_2,y,xlab=”x_1″,ylab=”x_2″,zlab=”y”,phi=30, surf=list(x = x_1.pred, y = x_2.pred, z = y.pred, facets = NA, fit = fitpoints)) #From other angles it is clear it is somewhat straight.
scatter3D(x_1,x_2,y,xlab=”x_1″,ylab=”x_2″,zlab=”y”,phi=60, surf=list(x = x_1.pred, y = x_2.pred, z = y.pred, facets = NA, fit = fitpoints)) #look at data. phi/theta is tilt.

Why do we need n-2? An example in R

Below is a simple example showing why we may want the $(\Sigma u^2_i )/ (n-2)$ as our estimates of $\large \sigma^2$, when our naive intuition may suggest we only want the simple average of squared errors $(\Sigma u^2_i )/ (n)$.

To show this in no uncertain terms, I have coded a linear regression by hand in R.  Also embedded in the work below are several rules I follow about writing code. They are rules 0-6.  There are many other rules, since code writing is an art.

####Coding in R
#### Rule 1: Always comment on every few lines of code. It is not unheard of to comment every single line, particularly for new coders, or complex code.
#### You will need to reference your work at a later date, and after about 3 months, the purpose is lost. Also, I need to read it.

#### Rule 2: Define your variables first. Luckily these names are shared for us.
#### For your projects, use names which are clear for your research: (y=crime in Williamsburg, VA, X= Number of jello puddings consumed)

set.seed(1223) #ensures replication. Sets seed of random number generators.
n<-25 #number of samples
x<-2*rnorm(n) #Our x’s come from a random sampling of X’s.
b_0<-10
b_1<-3 #Those cursed jello puddings are associated with increased crime. Linear regression is supportive of association- not causation.
u<-rnorm(n) #We satisfy both independent mean and zero mean assumptions
y<-b_0+b_1*x+u #This is defining our true Y. The true relationship is linear.

plot(x,y) #Rule 0, really. Always check your data.

#### Rule 3: After definitions begin your second stage of work. Probably trimming existing data, etc. Do these in the order they were added.
hat_b_1<-sum( (x-mean(x)) * (y-mean(y)) ) / sum( (x-mean(x))^2 ) #Spaces between any parenthesized section of operations. We need to be able to see which parentheses are which.
hat_b_1 # Rule 4: Indent work which is conceptually subordinate. Indent more as needed. Four spaces=1 tab.
hat_b_0<-mean(y)-hat_b_1*mean(x)
hat_b_0 # Rule 5: Check your work as you go along. For our example, I got 9.89

abline(a=hat_b_0, b=hat_b_1, col=”red”) #let’s add a red line of best fit. And we must see how our plot looks. Repeat rule 0.

hat_y<-hat_b_0+hat_b_1*x
hat_u<-hat_y-y

plot(x,hat_u) # Let’s see our residuals
hist(hat_u) # Let’s see our histogram

#### Rule 6: Keep your final analysis as punchy and short as possible without sacrificing clarity.
#### The mean sum of the squared errors (usually unknown to us as researchers)
sigma_sq<-sum(u^2)/n #this is the value we’re trying to estimate
sigma_sq_naive<-sum(hat_u^2)/n #this is a naive estimation of it
sigma_sq_hat<-sum(hat_u^2)/(n-2) #this turns out to be more accurate, particularly in small samples. If n->infinity this goes away. Try it for yourself!

#R, is this assessment true? Is sig_sq_hat a better estimator of sig_sq than our naive estimator? Is it true we need the (-2)?
(sigma_sq-sigma_sq_naive) > (sigma_sq-sigma_sq_hat)

Here is one of several plots made by this code, showing a nice linear regression over the data:

Please don’t forget the derivation of why this is true!  This is simply some supportive evidence that it might be true.

An Example of Plotting Multiple Time Series (Stock Values) on a Graph in R

I am currently in the process of designing a portfolio to manage investments. While such programs are not best plastered over the internet, a few basic concepts about plotting can be displayed.  For example, I have created a rather appealing plot, which demonstrates how to plot series of multiple images in a single plot, shown below:

Code is below, including my process to detrend the data. The critical lines are in bold, highlighting the fact that you can use sample(colors()) to select from the body of colors at random. This is useful when you may have to generate many plots, potentially without greatly detailed manual supervision, and you are not demanding publication-quality color selection (which is plausible for personal investigative use).

#after obtaining closing prices, you should make sure you clean your inputs. Ensure you know why there are NA’s, or you will make a critical error of omission.

closeprice<-log(closeprice)
data<-closeprice[is.finite(rowSums(closeprice)),]

#first-difference

data<-diff(data, lag=1, differences=1)
data<-na.omit(data)

#Check for any remaining trends in data over and above the natural cyclical or time-trending motion of the stocks!
#Detrend based off of the bond, a necessary part of even a basic CAPM portfolio
xhat<-lm(data$TYX.Close~1)$coefficients
detrended<-data-xhat #also, norm.
plot(index(detrended),detrended[,1],type=”l”)
for(n in 2:N){

lines(index(detrended),detrended[,n], col=sample(colors(),size=1))

}

Music and Math

Many people claim there is a strong correlation between music and math.
Below, I demonstrate that the patterns in music are NOT well predicted by typical statistical approaches.

Methodology:
I have taken a MIDI file of Beethoven’s 5th, and analyzed the track using non-parametric estimation techniques. These techniques included panel data techniques, ARMA, and extensive non-parametric estimation techniques (polynomial and Fourier series to capture cyclical components). I then use the song’s notes and my estimation technique to create a forecast of following notes. I then play the “forecasted song”.

Result:
After listening, the “forecasted song” does does not well match the original. As a consequence, I can state that the mathematical techniques common to forecasting do not well predict a song.  Below are several attempts which I have highlighted:

Caveat:
The R-squared for these estimations are in fact VERY high, in the high 90’s. (Only few of the coefficients are significant, the data is clearly overfitted in some regressions.) This song in fact falls into the so-called uncanny valley, and is only slightly deviant from the actual Beethoven’s 5th. However, the ear is strongly cultured to perfection in the subject of music, and the errors are devastating to us.

A Brief Presentation for the Student Investment Club (SIC)

In an attempt to branch out and see what other people do in terms of work, I’ve been creating a model for the Student Investment Club to simultaneously forecast GDP, CPI, and Unemployment.  While such a prediction is clearly overly ambitious for a casual effort, I made an attempt at it using some basic methodologies.  The dependent variables that I used in this case were guided by the preferences of the group, rather than by any particular theoretical model.  As such, I have very little faith in the model to be a powerful predictor on a fundamental level, but I do expect it to be correlated with the actual values.

Attached is my presentation (5-10 minutes) about my preliminary forecasts and findings. It is meant to be delivered to a nontechnical audience and is meant to be a partial (but not complete) disclosure of the problems with my approach. Below is a version of the model I am using to make such a forecast, with some admittedly sparse commentary.

SIC Presentation – Professional (Warm)

library(tseries)
library(quantmod)
library(systemfit)

# Old work functions as a great key
# gdp<-getSymbols(‘GDPC1′,src=’FRED’)
# indp<-getSymbols(‘INDPRO’, src=’FRED’)
# ism<-getSymbols(‘NAPM’, src=’FRED’)
# cap<-getSymbols(‘TCU’,src=’FRED’)
# wage<-getSymbols(‘AHETPI’,src=’FRED’) #Productivity? Proxy:wages
# ppi<-getSymbols(‘PPIACO’,src=’FRED’)
# unemploy<-U1RATENSA
# libor<-USDONTD156N
#
# cpi<-getSymbols(‘CPIAUCSL’,src=’FRED’)
# nom_pers_inc<-getSymbols(‘PINCOME’,src=’FRED’) #this might need to be real
# senti<-getSymbols(‘UMCSENT’,src=’FRED’)
# #demand<-getSymbols(‘DEMOTHCONS’,src=’FRED’)#Consumer demand? Proxy: request for more loans
# #cpi<-getSymbols(‘TCU’,src=’FRED’) #Total sales? Proxy: Change in buisness inventories

#Get the data
out<-NULL
b_names<-c(“GDPC1″,”INDPRO”,”NAPM”,”TCU”,”AHETPI”,”PPIACO”,”CPIAUCSL”,”PINCOME”,”UMCSENT”,”FEDFUNDS”,”U1RATENSA”)
getSymbols(b_names,src=’FRED’)
b<-list(GDPC1,INDPRO,NAPM,TCU,AHETPI,PPIACO,CPIAUCSL,PINCOME,UMCSENT,FEDFUNDS,U1RATENSA)
FEDFUNDS<-na.exclude(FEDFUNDS)
out<-lapply(b, aggregate, by=as.yearqtr, mean)

# Scale it appropriately.
series<-lapply(out,window,start=as.yearqtr(“2000 Q1”), end=as.yearqtr(“2013 Q1”))#trims to a consistant window.
series<-lapply(series,cbind)
series<-data.frame(series)
names(series)<-b_names
series<-log(series) #log the series
series<-as.ts(series) #need time series for this following operator:
series<-diff.ts(series[,c(“GDPC1″,”INDPRO”,”TCU”,”AHETPI”,”PPIACO”,”CPIAUCSL”,”PINCOME”,”UMCSENT”,”FEDFUNDS”,”U1RATENSA”)]) #first difference
lagGDP<-series[,”GDPC1″]
lagCPI<-series[,”CPIAUCSL”]
lagUNEMP<-series[,”U1RATENSA”]
series<-data.frame(series) #back to df
series$NAPM<-matrix(NAPM[(dim(NAPM)[1]+2-dim(series)[1]):dim(NAPM)[1]]) #Some may be stationary! series$lvl_UMCSENT<-matrix(UMCSENT[(dim(UMCSENT)[1]+2-dim(series)[1]):dim(UMCSENT)[1]])
series$lvl_TCU<-matrix(TCU[(dim(TCU)[1]+2-dim(series)[1]):dim(TCU)[1]]) series$lvl_NAPM<-matrix(NAPM[(dim(NAPM)[1]+2-dim(series)[1]):dim(NAPM)[1]])
series$lvl_FEDFUNDS<-matrix(FEDFUNDS[(dim(FEDFUNDS)[1]+2-dim(series)[1]):dim(FEDFUNDS)[1]]) series$t.index<-zooreg(series, start=as.yearqtr(“2000 Q1”),end=as.yearqtr(“2013 Q1″), frequency = 4) #need a time trend
series$quarter<-as.vector(seq(from=1,to=4, by=1)) # series$PINCOME_2<-(series$PINCOME)^2 #are these acceptable? # series$GDPC_2<-(series$GDPC1)^2 series_hold<-data.frame(series) # documentation http://cran.r-project.org/web/packages/systemfit/vignettes/systemfit.pdf series$Lead_GDPC1<-lag(zoo(lagGDP),k=+2, na.pad=TRUE)
series$Lead_CPIAUCSL<-lag(zoo(lagCPI),k=+2, na.pad=TRUE) series$Lead_U1RATENSA<-lag(zoo(lagUNEMP),k=+2, na.pad=TRUE) #impact takes at least 2 quarters. This is needed because we are missing CPI numbers for last quarter. Sentiment is delayed 6 months as propietary information. If it is set to +2, the estimates are to see what it would be like if we had the current info (pay for it).
eq1<- Lead_GDPC1 ~ INDPRO + lvl_NAPM + lvl_UMCSENT + GDPC1 + TCU + CPIAUCSL + FEDFUNDS + U1RATENSA + factor(quarter)
eq2<- Lead_CPIAUCSL ~ INDPRO + lvl_NAPM + lvl_UMCSENT + GDPC1 + TCU + CPIAUCSL + FEDFUNDS + U1RATENSA + factor(quarter)
eq3<- Lead_U1RATENSA ~ INDPRO + lvl_NAPM + lvl_UMCSENT + GDPC1 + TCU + CPIAUCSL + FEDFUNDS + U1RATENSA + factor(quarter)
eqsystem<-list(GDP=eq1,CPI=eq2,UNEMP=eq3)
# series<-data.frame(series)
fit<-systemfit(eqsystem, method=”SUR”, data=series)
pred<-predict(fit,series, se.pred=TRUE)
pred_ci<-predict(fit, series, interval=”confidence”, level=0.95) #note events are not normal.
plot(series$GDPC1, type=”l”, col=”darkgreen”, ylab=”% Change in GDP”, xlab=”Quarters (since 2000)”, main=”GDP forecast”) #the dimseries -40 gets me 10years. points(pred[1], type=”l”, col=”blue”, lty=5) points(pred_ci[,c(3)],type=”l”, col=”red”, lty=2) points(pred_ci[,c(2)],type=”l”, col=”red”, lty=2) legend(x=”bottomleft”,c(“Green= Actual GDP”,”Red= 95% CI”,”Blue=Forecast”), cex=0.90) plot(series$CPIAUCSL, type=”l”, col=”darkgreen”, ylab=”% Change in CPI”, xlab=”Quarters (since 2000)”,main=”CPI forecast”)
points(pred[3], type=”l”, col=”blue”, lty=5)
points(pred_ci[,5],type=”l”, col=”red”, lty=2)
points(pred_ci[,6],type=”l”, col=”red”, lty=2)
legend(x=”bottomleft”,c(“Green= Actual GDP”,”Red= 95% CI”,”Blue=Forecast”), cex=0.90)

plot(series\$U1RATENSA, type=”l”, col=”darkgreen”, ylab=”% Change in UNEMP”, xlab=”Quarters (since 2000)”, main=”UNEMP forecast”)
points(pred[5], type=”l”, col=”blue”, lty=5)
points(pred_ci[,8],type=”l”, col=”red”, lty=2)
points(pred_ci[,9],type=”l”, col=”red”, lty=2)
legend(x=”bottomleft”,c(“Green= Actual GDP”,”Red= 95% CI”,”Blue=Forecast”), cex=0.90)
summary(fit)

tail(pred)
pred<-rbind(0,rbind(0,pred))
pred_ci<-rbind(0,rbind(0,pred_ci))
tail(series[c(“CPIAUCSL”,”GDPC1″,”U1RATENSA”)])