-
Notifications
You must be signed in to change notification settings - Fork 2
/
cesBoundaryModels.R
189 lines (170 loc) · 7.49 KB
/
cesBoundaryModels.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
#' Fits CES boundary models
#'
#' The boundary models are given in Table 2 of
#' Heun, et al "An Empirical Investigation of the Role of Energy in Economic Growth"
#'
#' @param formula a CES formula in the form `y ~ a + b + c + d + time`
#' @param data historical time series data
#' @param nest a permutation of the integers 1 to the number of factors describing how to nest and permute
#' the variables in the formula.
#' @return a list of boundary models.
#' @examples
#' if (require(EconData) & require(dplyr)) {
#' cesBoundaryModels(iGDP ~ iK + iL + iQp + iYear, data=Calvin %>% filter(Country=="US"), nest=c(1,2,3))
#' cesBoundaryModels(iGDP ~ iK + iL + iQp + iYear, data=Calvin %>% filter(Country=="US"), nest=c(2,3,1))
#' cesBoundaryModels(iGDP ~ iK + iL + iYear, data=Calvin %>% filter(Country=="US"), nest=c(1,2))
#' }
#' @export
cesBoundaryModels <- function(formula, data, nest, method="nlm", subset=TRUE){
# f <- formula # to avoid problems while renaming is happening.
# use same data for all models, even if some models could make use of more complete data.
d <- subset(data, select = intersect(all.vars(formula), names(data)))
sdata <- data[complete.cases(d), unique(c(intersect(all.vars(formula), names(data)), names(data)))]
numFactors <- cesParseFormula(formula, nest)$numFactors
numFactorsInFormula <- cesParseFormula(formula, nest)$numFactorsInFormula
if ( ! numFactors %in% 2:3 ) {
stop("model must have 2 or 3 factors.")
}
if (numFactorsInFormula >=3) {
formulaRosetta <- list(
y = formula[[2]],
x1 = formula[[3]][[2]][[2]][[2]],
x2 = formula[[3]][[2]][[2]][[3]],
x3 = formula[[3]][[2]][[3]],
time = formula[[3]][[3]]
)
# pad to nestPlus using 1 + 2 + 3 = 6 to add missing index to end of the line
# when there are 2 vals in nest and 3 factors in formula.
if (length(nest) == 2) {
nestPlus <- c(nest, 6 - sum(nest))
} else {
nestPlus <- nest
}
# permute to reflect nesting
# when shuffling names rather than values, the permuation works "backwards"
# -- use order() to invert permuation. (Working on #242.)
#
names(formulaRosetta)[2:4] <- c("x1", "x2", "x3")[order(nestPlus)]
} else {
formulaRosetta <- list(
y = formula[[2]],
x1 = formula[[3]][[2]][[2]],
x2 = formula[[3]][[2]][[3]],
time = formula[[3]][[3]]
)
# permute to reflect nesting -- see above for comment about order()
names(formulaRosetta)[2:3] <- c("x1", "x2")[order(nest)]
}
# old names - new names
formulaTemplates <-
list(
"01:1-1-" = log(y) - log(x1) ~ time,
"02:0-1-" = log(y) - log(x2) ~ time,
"03:--0-" = log(y) - log(x3) ~ time,
"04:-01-" = log(y) - log(pmin(x1, x2)) ~ time,
"05:1--0" = log(y) - log(pmin(x1, x3)) ~ time,
"06:0--0" = log(y) - log(pmin(x2, x3)) ~ time,
"07:*N1-" = log(y) - log(delta_1*x1 + (1-delta_1)*x2) ~ time,
"08:1-*N" = log(y) - log(delta*x1 + (1-delta)*x3) ~ time,
"09:0-*N" = log(y) - log(delta*x2 + (1-delta)*x3) ~ time,
"13:-0-0" = log(y) - log(pmin(x1, x2, x3)) ~ time,
"14:-0*N" = log(y) - log( delta*pmin(x1, x2) + (1-delta)*x3 ) ~ time,
"15:*N-0" = log(y) - log( pmin(delta_1*x1 + (1-delta_1)*x2, x3)) ~ time,
"16:*N*N" = log(y) - log( delta*(delta_1*x1 + (1-delta_1)*x2) + (1-delta)*x3) ~ time,
"18:**-0" = log(y) - log( pmin((delta_1*x1^(-rho_1) + (1-delta_1)*x2^(-rho_1))^(-1/rho_1), x3) ) ~ time,
"19:***N" = log(y) - log( (delta * (delta_1*x1^(-rho_1) + (1-delta_1)*x2^(-rho_1))^(-1/rho_1) + (1-delta)*x3 ) ) ~ time,
"20:*N**" = log(y) - log( (delta * (delta_1*x1 + (1-delta_1)*x2)^(-rho) + (1-delta)*x3^(-rho) ) ^ (-1/rho)) ~ time,
"10:**1-" = y ~ x1 + x2 + x3 + time,
"11:1-**" = y ~ x1 + x2 + x3 + time,
"12:0-**" = y ~ x1 + x2 + x3 + time,
"17:-0**" = y ~ pmin(x1, x2) + placeholder + x3 + time
)
# Values that don't appear in the model can be set to their fixed values and nlm()
# will leave them alone.
# But nlm() doesn't like Inf even if the parameter isn't used, so when rho or rho_1 is Inf, we
# set sigma or sigma_1, respectively, to 0 instead of setting rho or rho_1 to Inf.
# This will cost some effort down stream to recover the rho values. See makeNatCoef.
# There may be a better solution, but this will get us going for the moment.
params <- list(
"01:1-1-" = c(delta_1 = 1, delta=1),
"02:0-1-" = c(delta_1 = 0, delta=1),
"03:--0-" = c(delta = 0),
"04:-01-" = c(sigma_1 = 0, delta=1),
"05:1--0" = c(sigma = 0, delta_1 = 1.0),
"06:0--0" = c(sigma = 0, delta_1 = 0.0),
"07:*N1-" = c(delta_1 = 0.5, delta = 1.0, rho_1 = -1),
"08:1-*N" = c(delta = 0.5, delta_1= 1.0, rho = -1),
"09:0-*N" = c(delta = 0.5, delta_1 = 0.0, rho = -1),
# 5
# 15
# 16
"13:-0-0" = c(sigma = 0, sigma_1 = 0.0),
"14:-0*N" = c(delta = 0.5, sigma_1 = 0.0, rho = -1),
"15:*N-0" = c(delta_1 = 0.5, rho_1 = -1.0, sigma = 0),
"16:*N*N" = c(delta = 0.5, delta_1 = 0.5, rho_1 = -1, rho = -1),
# 18
"18:**-0" = c(delta_1 = 0.5, rho_1 = 0.25, sigma = 0.0),
"19:***N" = c(delta = 0.5, delta_1 = 0.5, rho_1 = 0.25, rho = -1),
"20:*N**" = c(delta = 0.5, delta_1 = 0.5, rho = 0.25, rho_1 = -1)
)
formulas <-
lapply(
formulaTemplates,
function(ft)
as.formula(do.call(substitute, list( ft, formulaRosetta) ) )
)
keep <- sapply(formulaTemplates, function(ft) numFactors >=3 | !("x3" %in% all.vars(ft)))
plmModel <- sapply(1:length(formulaTemplates), function(x) x <= length(params)) # params for plm() only
# fit all the models
plmModels <-
Map(
function(formula, param) {
res <- eval(substitute( plm(formula, data = sdata, param = param, method = method),
list(formula=formula, param=param, method=method)))
# for all of our models, log(gamma) and lambda are first two coefficients
if (! is.null(res)) {
names(res$coefficients)[1] <- "logscale"
names(res$coefficients)[2] <- "lambda"
}
res$nest = nest
res
},
formulas[keep & plmModel & subset], params[keep & plmModel & subset]
)
cesModels <-
Map(
function(formula, nest) {
eval(substitute(
cesModel(formula, data = sdata, nest = nest, constrained = TRUE, fitBoundaries = FALSE),
list(formula=formula)
))
},
formulas[keep & !plmModel & (1:20 < 20) & subset], # remove ugly case (20)
list(c(1,2), c(1,3), c(2,3))[any(keep & !plmModel & (1:20 < 20) & subset)] # adjust nest to leave one out
)
# Now handle the ugly case. Since cesEst() can only work with variables, we need
# to compute a variable and add it to the data frame before calling cesModel.
if(keep[20]) {
tryCatch( {
formulaPmin <- formulas[[20]]
pminVar <- deparse(formulaPmin[[3]][[2]][[2]][[2]])
sdata[[pminVar]] <- eval(formulaPmin[[3]][[2]][[2]][[2]], sdata)
cesModels <-
c(cesModels,
list("17:-0**" =
eval(substitute(
cesModel(f, data=sdata, nest = c(1,3), constrained=TRUE, fitBoundaries=FALSE),
list(f = formulaPmin)
))
)
)
}, error = function(e) warning(e)
)
}
o <- order( names(formulaTemplates[keep]) )
return(
Map(function(model, bd) { model$boundary <- bd; model },
model = c(plmModels, cesModels)[o],
bd = names(c(plmModels, cesModels)[o])
) )
}