-
Notifications
You must be signed in to change notification settings - Fork 0
/
merge_and_analysis.R
290 lines (244 loc) · 16.7 KB
/
merge_and_analysis.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
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
## Connor P. Jackson second year paper
## Merge and analyze the data
library(data.table)
library(bit64)
library(lubridate)
library(tidyr)
library(DescTools)
library(tidycensus)
## ---- import-data ----
nchomes <- fread("../data_buyouts/ZAsmt_NC.csv", key="ImportParcelID", index="latest")
nctrans <- fread("../data_buyouts/ZTrans_NC.csv", key="ImportParcelID", na.strings = "")
source("NFIP_claims_processing.R")
nchomes <- nchomes[!is.na(censusTract)]
# TODO: recreate missing census tracts from zipcodes or latlongs
# Remove unreasonable years built, cross-apply missing years built, drop properties with unknown years
nchomes[EffectiveYearBuilt < YearBuilt, EffectiveYearBuilt := NA]
nchomes[EffectiveYearBuilt > year(Sys.time()), EffectiveYearBuilt := NA]
nchomes[is.na(EffectiveYearBuilt), EffectiveYearBuilt := YearBuilt]
nchomes[is.na(YearBuilt), YearBuilt := EffectiveYearBuilt]
nchomes <- nchomes[!is.na(YearBuilt)]
## Handle multiple assessment records in a year. Keep the most recent one
nchomes[, record_date := my(ExtractDate)]
nchomes[, record_year := year(record_date)]
setorder(nchomes, ImportParcelID, record_date)
nchomes <- unique(nchomes, by=c("ImportParcelID", "record_year"), fromLast=TRUE)
sfha_homes <- nchomes[SFHA_TF == TRUE]
rm(nchomes)
modal_val <- function(confusingVar) {
if (length(confusingVar) == uniqueN(confusingVar)) return(tail(confusingVar, n = 1))
return(tail(Mode(confusingVar), n = 1))
}
## TODO: handle changes in census tracts better
## If a house is listed in different tracts in different years, take the modal tract. For ties, take the later
sfha_homes[, censusTract := as.integer64(modal_val(as.numeric(as.character(censusTract)))), by = .(ImportParcelID)]
# Do the same thing for year built
sfha_homes[, YearBuilt := modal_val(YearBuilt), by = .(ImportParcelID)]
sfha_homes[, EffectiveYearBuilt := modal_val(EffectiveYearBuilt), by = .(ImportParcelID)]
## ---- CH-create-property-long-data ----
# Define the transaction year
nctrans[, date := DocumentDate]
nctrans[is.na(date), date := RecordingDate]
nctrans[, year := year(date)]
nctrans[, IntraFamilyTransferFlag := IntraFamilyTransferFlag == "Y"]
nctrans[, TransferTaxExemptFlag := TransferTaxExemptFlag == "Y"]
nctrans[DocumentTypeStndCode == "", DocumentTypeStndCode := NA]
nctrans[LoanTypeStndCode == "", LoanTypeStndCode := NA]
nctrans[SalesPriceAmountStndCode == "", SalesPriceAmountStndCode := NA]
nctrans[, mortgage_nfip_req := LoanTypeProgramStndCode != ""]
dataclasses <- c(NA, "", "D", "H", "M", "P")
doctypes <- c(NA, "", "ADDE","AFDV","AGSL","BFDE","BSDE","CDDE","CHDE","CMDE","CPDE","CTSL",
"DEED","EXCH","GRDE","IDDE","LWDE","MTGE","NOWR","QCDE","RCDE","SPWD","WRDE")
pricecodes <- c(NA, "", "AF","AV","CF","CM","CN","CR","CS","CU","EP","HB","LN","MP","RA","RD","ST")
loantypes <- c(NA, "", "AC","AS","BL","CE","CM","CL","DP","FO","FM","FA","PM","SL")
# identify the arms length transactions, not intra-family transfers, refinances, foreclosures, etc,
# only for the homes in Special Flood Hazard Areas
nctrans_armslen <- nctrans[(ImportParcelID %in% sfha_homes[, unique(ImportParcelID)] &
DataClassStndCode %in% dataclasses & DocumentTypeStndCode %in% doctypes &
SalesPriceAmountStndCode %in% pricecodes & LoanTypeStndCode %in% loantypes &
mortgage_nfip_req == TRUE & IntraFamilyTransferFlag == FALSE & year >= 1995)]
# Define only a single transaction per property per year. If there are multiple transaction
# records in a year, choose the one with the highest (non-missing) recorded sale price
setkey(nctrans_armslen, ImportParcelID, year)
setorder(nctrans_armslen, ImportParcelID, year, -SalesPriceAmount, na.last = TRUE)
nctrans_armslen <- unique(nctrans_armslen, by = c("ImportParcelID", "year"))
nctrans_armslen[, transaction_obs := TRUE]
# construct the panel dataset by filling in every parcel x year combination
panel_frame <-CJ(ImportParcelID = sfha_homes[, unique(ImportParcelID)], year = nctrans_armslen[, unique(year)])
keepvars <- c("ImportParcelID", "date", "year", "SalesPriceAmount", "mortgage_nfip_req", "transaction_obs")
nctrans_panel <- merge(nctrans_armslen[, ..keepvars], panel_frame, all = TRUE)
rm(nctrans_armslen, panel_frame, nctrans)
## We apply the assessor data onto the transaction panel, assuming the old assessment records
## hold until a new one occurs
nctrans_panel[, roll_year := year]
setkey(nctrans_panel, ImportParcelID, roll_year)
sfha_homes[, roll_year := record_year]
setkey(sfha_homes, ImportParcelID, roll_year)
nctrans_panel <- sfha_homes[nctrans_panel, roll = TRUE, rollends = TRUE]
nctrans_panel[is.na(transaction_obs), transaction_obs := FALSE]
nctrans_panel[, transaction_obs := as.integer(transaction_obs)]
nctrans_panel[, roll_year := NULL]
setnames(nctrans_panel, "FLD_ZONE", "floodZone")
setkey(nctrans_panel, censusTract, floodZone, YearBuilt, ImportParcelID, year)
## ---- flood-events ----
# Start the flood year August 1
claims[, year := year(yearofLoss)]
claims[, year_corrected := year(yearofLoss)]
claims[month(dateOfLoss) >= 8, year_corrected := year_corrected + 1]
flood_frame <- CJ(tract_zone = unique(c(sfha_homes[, unique(paste(censusTract, FLD_ZONE, sep="_"))],
claims[sfha == TRUE, unique(paste(censusTract, floodZone, sep="_"))])),
year = claims[, unique(year(yearofLoss))])
flood_count <- claims[sfha == TRUE, .(flood_count = .N), by = .(tract_zone = paste(censusTract, floodZone, sep="_"), year)]
setkey(flood_count, tract_zone, year)
flood_panel <- merge(flood_count, flood_frame, all = TRUE)
rm(flood_count, flood_frame)
flood_panel[, c("censusTract", "floodZone") := tstrsplit(tract_zone, "_", fixed = TRUE, type.convert = TRUE)]
flood_panel[, censusTract := as.integer64(censusTract)]
flood_panel[is.na(flood_count), flood_count := 0]
flood_panel[, flood_event := as.integer(flood_count > 0)]
flood_panel[, tract_zone := NULL]
## TODO: change years to start in August or at the start of flood season?
## ---- nfip-policy-holders ----
policies_frame <- CJ(tract_zone_year = unique(c(sfha_homes[, unique(paste(censusTract, FLD_ZONE, YearBuilt, sep="_"))],
policies[sfha == TRUE, unique(paste(censusTract, floodZone, originalConstructionDate, sep="_"))])),
year = policies[, unique(policyyear)])
policies_status <- policies[sfha == TRUE, .(policies_count = .N, postFIRM = (mean(postFIRMConstructionIndicator) > 0.5)),
by = .(tract_zone_year = paste(censusTract, floodZone, originalConstructionDate, sep="_"),
year = policyyear)]
setkey(policies_status, tract_zone_year, year)
policies_panel <- merge(policies_status, policies_frame, all = TRUE)
policies_panel[, c("censusTract", "floodZone", "originalConstructionDate") := tstrsplit(tract_zone_year,
"_", fixed = TRUE, type.convert = TRUE)]
policies_panel[, tract_zone_year := NULL]
policies_panel[, censusTract := as.integer64(censusTract)]
policies_panel[is.na(policies_count), policies_count := 0]
rm(policies_frame, policies_status)
# TODO: Find other data on year of initial FIRM (and maybe CRS class as well?)
FIRMdates <- policies_panel[postFIRM == TRUE, .(FIRMyear = min(originalConstructionDate, na.rm = TRUE)),
keyby = .(censusTract, floodZone)]
FIRMdates[FIRMyear < 1974, FIRMyear := 1974]
policies_panel[, postFIRM := NULL]
setnames(policies_panel, "originalConstructionDate", "YearBuilt")
setkey(policies_panel, censusTract, floodZone, YearBuilt, year)
## ---- combine-panels ----
tzy_panel <- nctrans_panel[, .(properties_count = .N, TotalAssessedValue = mean(TotalAssessedValue),
TotalMarketValue = mean(TotalMarketValue),
transaction_count = sum(transaction_obs), SalesPriceAmount = mean(SalesPriceAmount)),
keyby = .(censusTract, floodZone, YearBuilt, year)]
tzy_panel[, transaction_prob := transaction_count / properties_count]
tzy_panel <- merge(tzy_panel, policies_panel, all.x = TRUE)
tzy_panel <- merge(tzy_panel, flood_panel, by = c("censusTract", "floodZone", "year"), all.x = TRUE)
tzy_panel <- merge(tzy_panel, FIRMdates, by = c("censusTract", "floodZone"), all.x = TRUE)
setkey(tzy_panel, censusTract, floodZone, YearBuilt, year)
tzy_panel[, policy_prob := policies_count / properties_count]
tzy_panel[, adapted := YearBuilt > FIRMyear]
tzy_panel[YearBuilt < 1974, adapted := FALSE]
tzy_panel[YearBuilt == FIRMyear, adapted := NA] # ambiguous within-year FIRM timing
tzy_panel[, adapted := as.integer(adapted)]
tzy_panel[, adapted_text := ifelse(adapted == 1, "adapted (new homes)", "non-adapted (old homes)")]
tzy_panel[, reg_reform := as.integer(year >= 2013)]
# # Create the lag flood policy and flood events data
# setorder(tzy_panel, censusTract, floodZone, YearBuilt, year)
# tzy_panel[, policies_L1 := shift(policies_count, 1, type = "lag"), by = .(censusTract, floodZone, YearBuilt)]
# tzy_panel[, policies_L2 := shift(policies_count, 2, type = "lag"), by = .(censusTract, floodZone, YearBuilt)]
# tzy_panel[, policies_L3 := shift(policies_count, 3, type = "lag"), by = .(censusTract, floodZone, YearBuilt)]
# tzy_panel[, policies_F1 := shift(policies_count, 1, type = "lead"), by = .(censusTract, floodZone, YearBuilt)]
# tzy_panel[, flood_L1 := shift(flood_event, 1, type = "lag"), by = .(censusTract, floodZone, YearBuilt)]
# tzy_panel[, flood_L2 := shift(flood_event, 2, type = "lag"), by = .(censusTract, floodZone, YearBuilt)]
# tzy_panel[, flood_L3 := shift(flood_event, 3, type = "lag"), by = .(censusTract, floodZone, YearBuilt)]
# tzy_panel[, flood_F1 := shift(flood_event, 1, type = "lead"), by = .(censusTract, floodZone, YearBuilt)]
# tzy_panel[, policy_prob_L1 := policies_L1 / properties_count]
# tzy_panel[, policy_prob_L2 := policies_L2 / properties_count]
# tzy_panel[, policy_prob_L3 := policies_L3 / properties_count]
# tzy_panel[, policy_prob_F1 := policies_F1 / properties_count]
# tzy_panel[, reg_reform_L1 := shift(reg_reform, 1, type = "lag"), by = .(censusTract, floodZone, YearBuilt)]
# tzy_panel[, reg_reform_L2 := shift(reg_reform, 2, type = "lag"), by = .(censusTract, floodZone, YearBuilt)]
# tzy_panel[, reg_reform_L3 := shift(reg_reform, 3, type = "lag"), by = .(censusTract, floodZone, YearBuilt)]
# tzy_panel[, reg_reform_F1 := shift(reg_reform, 1, type = "lead"), by = .(censusTract, floodZone, YearBuilt)]
#
# # tractable instruments: legislative changes (2013 x adapted)
# # construct the instruments manually: flood_event:policy_prob | flood_L1:policy_prob_L1 | flood_L2:policy_prob_L2
# tzy_panel[, flooded_insured := flood_event*policy_prob]
# tzy_panel[, flooded_insured_L1 := flood_L1*policy_prob_L1]
# tzy_panel[, flooded_insured_L2 := flood_L2*policy_prob_L2]
# tzy_panel[, flooded_insured_L3 := flood_L3*policy_prob_L3]
# tzy_panel[, flooded_insured_F1 := flood_F1*policy_prob_F1]
#
# year range: 2009-2016 (2017 is only thru September) (2012-2016 with lags)
tzy_panel[, in_sample := ((year %in% 2009:2016) & !is.na(adapted) & YearBuilt < 2009)]
tzy_panel[, sample_2L := in_sample & year > 2010]
tzy_panel[, sample_3L := in_sample & year > 2011]
tzy_panel[, panel_id := as.factor(paste(censusTract, floodZone, sep="_"))]
tzy_panel[, censusTract := as.factor(censusTract)]
tzy_panel[, YearBuilt := as.factor(YearBuilt)]
tzy_panel[, year := as.factor(year)]
# TODO: figure out issue with policies exceeding properties
bad_tracts <- tzy_panel[in_sample == TRUE & policies_count > properties_count, unique(panel_id)]
tzy_panel[panel_id %in% bad_tracts, in_sample := FALSE]
## ---- house-level-panel ----
nctrans_panel[, properties_count := .N, by = .(censusTract, floodZone, YearBuilt, year)]
nctrans_panel <- merge(nctrans_panel, policies_panel, by = c("censusTract", "floodZone", "YearBuilt", "year"), all.x = TRUE)
nctrans_panel <- merge(nctrans_panel, flood_panel, by = c("censusTract", "floodZone", "year"), all.x = TRUE)
nctrans_panel <- merge(nctrans_panel, FIRMdates, by = c("censusTract", "floodZone"), all.x = TRUE)
nctrans_panel[, policy_prob := policies_count / properties_count]
nctrans_panel[, adapted := YearBuilt > FIRMyear]
nctrans_panel[YearBuilt < 1974, adapted := FALSE]
nctrans_panel[YearBuilt == FIRMyear, adapted := NA] # ambiguous within-year FIRM timing
nctrans_panel[, adapted := as.integer(adapted)]
nctrans_panel[, adapted_text := ifelse(adapted == 1, "adapted (new homes)", ifelse(is.na(adapted), NA, "non-adapted (old homes)"))]
nctrans_panel[, nonadapted := 1-adapted]
nctrans_panel[, reg_reform := as.integer(year >= 2013)]
# Create the lead transaction for the DD approach
setorder(nctrans_panel, censusTract, floodZone, YearBuilt, ImportParcelID, year)
nctrans_panel[, trans_F1 := shift(transaction_obs, 1, type = "lead"), by = .(censusTract, floodZone, YearBuilt, ImportParcelID)]
nctrans_panel[, trans_F2 := shift(transaction_obs, 2, type = "lead"), by = .(censusTract, floodZone, YearBuilt, ImportParcelID)]
nctrans_panel[, sale_2years := apply(.SD, 1, any), .SDcols = c("transaction_obs", "trans_F1", "trans_F2")]
# create the lag other variables for the event study approach
nctrans_panel[, policies_L1 := shift(policies_count, 1, type = "lag"), by = .(censusTract, floodZone, YearBuilt, ImportParcelID)]
nctrans_panel[, policies_L2 := shift(policies_count, 2, type = "lag"), by = .(censusTract, floodZone, YearBuilt, ImportParcelID)]
nctrans_panel[, policies_L3 := shift(policies_count, 3, type = "lag"), by = .(censusTract, floodZone, YearBuilt, ImportParcelID)]
nctrans_panel[, policies_F1 := shift(policies_count, 1, type = "lead"), by = .(censusTract, floodZone, YearBuilt, ImportParcelID)]
nctrans_panel[, flood_L1 := shift(flood_event, 1, type = "lag"), by = .(censusTract, floodZone, YearBuilt, ImportParcelID)]
nctrans_panel[, flood_L2 := shift(flood_event, 2, type = "lag"), by = .(censusTract, floodZone, YearBuilt, ImportParcelID)]
nctrans_panel[, flood_L3 := shift(flood_event, 3, type = "lag"), by = .(censusTract, floodZone, YearBuilt, ImportParcelID)]
nctrans_panel[, flood_F1 := shift(flood_event, 1, type = "lead"), by = .(censusTract, floodZone, YearBuilt, ImportParcelID)]
nctrans_panel[, policy_prob_L1 := policies_L1 / properties_count]
nctrans_panel[, policy_prob_L2 := policies_L2 / properties_count]
nctrans_panel[, policy_prob_L3 := policies_L3 / properties_count]
nctrans_panel[, policy_prob_F1 := policies_F1 / properties_count]
nctrans_panel[, reg_reform_L1 := shift(reg_reform, 1, type = "lag"), by = .(censusTract, floodZone, YearBuilt, ImportParcelID)]
nctrans_panel[, reg_reform_L2 := shift(reg_reform, 2, type = "lag"), by = .(censusTract, floodZone, YearBuilt, ImportParcelID)]
nctrans_panel[, reg_reform_L3 := shift(reg_reform, 3, type = "lag"), by = .(censusTract, floodZone, YearBuilt, ImportParcelID)]
nctrans_panel[, reg_reform_F1 := shift(reg_reform, 1, type = "lead"), by = .(censusTract, floodZone, YearBuilt, ImportParcelID)]
# manually create the instrument
nctrans_panel[, flooded_insured := flood_event*policy_prob]
nctrans_panel[, flooded_insured_L1 := flood_L1*policy_prob_L1]
nctrans_panel[, flooded_insured_L2 := flood_L2*policy_prob_L2]
nctrans_panel[, flooded_insured_L3 := flood_L3*policy_prob_L3]
nctrans_panel[, flooded_insured_F1 := flood_F1*policy_prob_F1]
# year range: 2009-2016 (2017 is only thru September)
nctrans_panel[, full_sample := ((year %in% 2009:2016) & !is.na(adapted) & YearBuilt < 2009)]
nctrans_panel[, panel_id := as.factor(paste(censusTract, floodZone, sep="_"))]
nctrans_panel[, censusTract := as.factor(censusTract)]
nctrans_panel[, YearBuilt_f := as.factor(YearBuilt)]
nctrans_panel[, year_f := as.factor(year)]
# TODO: figure out issue with policies exceeding properties
# Most of the homes with mismatched policies are NEW homes! (~1/4 built 2000-2005)
# Possible issues: geocoding problems, misreporting of construction year in either
# dataset, should be using EffectiveYearBuilt rather than YearBuilt
bad_tracts <- nctrans_panel[full_sample == TRUE & policy_prob > 1, unique(panel_id)]
nctrans_panel[panel_id %in% bad_tracts, full_sample := FALSE]
# Construct the analysis samples for the diff in diff and the event studies
nctrans_panel[, dd_sample := full_sample & year <= 2014]
nctrans_panel[, es_sample_2L := full_sample & year > 2010]
nctrans_panel[, es_sample_3L := full_sample & year > 2011]
## ---- census-data ----
# v2009 <- load_variables(2009, "acs1", cache = TRUE)
# v2009_5 <- load_variables(2009, "acs5", cache = TRUE)
#
# acs_vars <- c("B01001_002", "B01002_001", "B01003_001", "B02001_002", "B02001_003", "B03001_003",
# "B17001_002", "B06011_001", "C18108_010", "C18130_016",
# "C18108_002", "C18130_002")
# acs_2009 <- get_acs(geography = "tract", variables = acs_vars, state = "NC", moe = 95,
# year = 2009, survey = "acs5")