Explore_RevGSP(3).R

# This script explore the relationship between GSP and state gov revenue and does exploratory modeling 
# It is also for the first draft of the policy brief. 


#**********************************************************************
#                          Notes                                   ####
#**********************************************************************

# To-do list for GSP and tax revenue section


# What to model:
# 1. Personal income tax: relationship with GSP growth and returns (stock and/or portfolio)
# 2. General Sales tax: relationship with GSP, how the relationship changes during recessions
# 3. Other taxes (non-PIT-non-sales taxes): relationship with GSP
# 4. Construct stylized government types 

# Note: 
#  - Use real values for now. When integrated with macro model, which uses real GDP, 
#    add the assumed inflation rate to the GSP growth rate. So we do not model inflation 
#    in this version. 


# Other relationships to examine:
  # 1. GSP growth and total own soure revenue growth,
  # 2. GSP growth and total tax revenue growth
  # 3. GSP growth and personal income tax growth
  # 4. GSP growth and sales tax growth
  # 5. GSP growth and property tax growth
  # 6. GSP growth and corporate income tax growth
  # 7. GSP growth and growth of total tax revenue minus PIT and sales
  # 8. national GDP growth and GSP growth


# Notes on real and nominal variables
#  - results for nominal variables for 1995-2015 can be compared with Pew results
#  - GDP/GSP are modeled in real terms in the macro model, so the GDP-revenue relationship should be
#    modeld using real term. But note that pension contributions are modeled in nominal terms, which 
#    must be compared against nominal gov revenue. 
#  - Real revenue variables are in 2015 dollar and adjustment factor for inflation is the same across
#    all states, gov levels and types of tax, while real GSPs are in 2009 dollar and adjustment factors 
#    for inflation differ across states.


# How sales taxes are divided between state and local governments:
 # https://taxfoundation.org/state-and-local-sales-tax-rates-in-2017/
 # Five states do not have statewide sales taxes: Alaska, Delaware, Montana, New Hampshire, and Oregon. 
 # Of these, Alaska and Montana allow localities to charge local sales taxes.



#**********************************************************************
#                           Packages                               ####
#**********************************************************************

library(tidyverse)
library(broom)
library(readxl)
library(magrittr)
library(stringr)
library(forcats)
library(grid)
library(gridExtra)
library(scales)
library(knitr)

# packages for time series modeling
library(astsa)    # companion package
library(TSA)      # companion package;  arimax: flexible transfer function model
library(tseries)  #
library(forecast) # Arima
library(MSwM)
library(TTR)
library(dynlm)
library(broom)

#library(MSBVAR)

# packages for ts
library(zoo)
library(xts)


library(timetk)
library(tidyquant)

library(lubridate)
library(feather)

library(psych) # describe

options(tibble.print_max = 60, tibble.print_min = 60)


# check tidyquant, timetk, sweep (broom ), tibbletime
# Intro to zoo  cran.r-project.org/web/packages/zoo/vignettes/zoo-quickref.pdf
# sweep: http://www.business-science.io/code-tools/2017/07/09/sweep-0-1-0.html

#**********************************************************************
#                     Global settings and tools                    ####
#**********************************************************************
dir_data_raw <- "data_raw/"
dir_data_out <- "data_out/"
dir_fig_out <- "policyBrief_out/"


# NBER recession periods, post-WWII
recessionPeriods <- 
  matrix(c(
  	1953+2/4, 1954+2/4,
  	1957+3/4, 1958+2/4,
  	1960+2/4, 1961+1/4,
  	1969+4/4, 1970+4/4,
  	1973+4/4, 1975+1/4,
  	1980+1/4, 1980+3/4,
  	1981+3/4, 1982+4/4,
  	1990+3/4, 1991+1/4,
  	2001+1/4, 2001+4/4,
  	2007+4/4, 2009+2/4
  ) , ncol = 2, byrow = T) %>% 
  as.data.frame() %>% 
	rename(peak =   V1,
				 trough = V2) %>% 
	mutate(peak = peak - 1/4,
				 trough = trough - 1/4)


get_logReturn <- function(x){
	if(any(x <= 0, na.rm = TRUE)) stop("Nagative value(s)")
	log(x/lag(x))
}




# RIG colors and theme
RIG.blue  <- "#003598"
RIG.red   <- "#A50021"
RIG.green <- "#009900"
RIG.yellow <- "#FFFF66"
RIG.purple <- "#9966FF"
RIG.yellow.dark <- "#ffc829"
RIG.orange <- "#fc9272"

demo.color6 <- c(RIG.red,
								 RIG.orange,
								 RIG.purple,
								 RIG.green ,
								 RIG.blue,
								 RIG.yellow.dark)


RIG.theme <- function() {
	theme(
		panel.grid.major.x = element_blank(),
		panel.grid.minor.x = element_blank(),
		panel.grid.minor.y = element_blank(),
		panel.grid.major.y = element_line(size = 0.5, color = "gray80"),
		plot.title = element_text(hjust = 0.5),
		plot.subtitle = element_text(hjust = 0.5),
		plot.caption = element_text(hjust = 0, size = 9)
	)
}

RIG.themeLite <- function() {
	theme(
		plot.title = element_text(hjust = 0.5),
		plot.subtitle = element_text(hjust = 0.5),
		plot.caption = element_text(hjust = 0, size = 9)
	)
}




#**********************************************************************
#                       Loading Data                               ####
#**********************************************************************
# Loading saved data 
load(paste0(dir_data_out, "data_RevGSP.RData"))
load(paste0(dir_data_out, "dataAll.RData"))


# df_RevGSP loaded

# Variables in df_RevGSP
#  Indices: state, state_abb, year
#  GSP variables
#     - RGSP_SIC:   From BEA,  1963-1997, based on SIC
#     - RGSP_NAICS: From BEA,  1997-2016, based on NAICS
#     - NGSP:       From FRED, 1997-2016
#  Tax and revenue variables: 1977-2015
#     - suffix for gov levels: 
#        - local
#        - state
#        - SL: state and local
#     - suffix for real or nominal: real / nom
# 
#     - variables:   
#      urban code    Var name					  Var description
#        'R01',     'rev_tot',          'Total revenue', 
#        'R02',     'rev_tot_ownSrc',   'Total Rev-Own Sources',
#        'R03',     'rev_gen',          'General Revenue',
#        'R04',     'rev_gen_ownSrc',   'Gen Rev-Own Sources',
#        'R05',     'tax_tot',          'Total Taxes',
#        'R06',     'tax_property',     'Property Taxes',
#        'R08',     'tax_sales_tot',    'Tot Sales & Gr Rec Tax',
#        'R09',     'tax_sales_gen',    'Total Gen Sales Tax (T09)',
#        'R10',     'tax_sales_select', 'Total select Sales Tax',
#        'R27',     'tax_indivIncome',  'Individual Income Tax (T40)',
#        'R28',     'tax_corpIncome',   'Corp Net Income Tax',
#        'R36',     'chgs_Misc',        'Tot Chgs and Misc Rev'
#      - with missing values 

# Real terms (BEA real GSP starts from 1987)
df_real <- 
	df_RevGSP %>%
	select(state_abb, year, 
				 RGSP_SIC, RGSP_NAICS, 
				 tax_tot_real_state, 
				 tax_indivIncome_real_state,
				 tax_sales_tot_real_state,
				 tax_sales_gen_real_state,
				 tax_sales_select_real_state,
				 tax_corpIncome_real_state,
				 tax_property_real_state) %>% 
	mutate(tax_other_real_state = tax_tot_real_state - tax_indivIncome_real_state - 
				                        tax_sales_tot_real_state - tax_corpIncome_real_state - tax_property_real_state,
				 tax_nonPITsalesgen_real_state = tax_tot_real_state - tax_indivIncome_real_state - tax_sales_gen_real_state,
				 tax_nonPITsalestot_real_state = tax_tot_real_state - tax_indivIncome_real_state - tax_sales_tot_real_state,
				 RGSPc = ifelse(year >= 1997, RGSP_NAICS, RGSP_SIC),
				 RGSPc = 1000 * RGSPc,
				 PIT_tax = 100 * tax_indivIncome_real_state/tax_tot_real_state, 
				 tax_GSP = 100 * tax_tot_real_state / RGSPc)

df_dlreal <- 
	df_real %>% 
	group_by(state_abb) %>% 
	mutate_at(vars(-year), funs(100 * log(./lag(.)))) %>% 
	mutate(RGSP = ifelse(year > 1997, RGSP_NAICS, RGSP_SIC))

df_real
df_dlreal


# Nominal terms (BEA nominal GSP starts from 1963)
df_nom <- 
	df_RevGSP %>%
	select(state_abb, year, 
				 NGSP_SIC, NGSP_NAICS, 
				 tax_tot_nom_state, 
				 tax_indivIncome_nom_state,
				 tax_sales_tot_nom_state,
				 tax_sales_gen_nom_state,
				 tax_sales_select_nom_state,
				 tax_corpIncome_nom_state,
				 tax_property_nom_state) %>% 
	mutate(tax_other_nom_state = tax_tot_nom_state - tax_indivIncome_nom_state - 
				 	                     tax_sales_tot_nom_state - tax_corpIncome_nom_state - tax_property_nom_state,
				 tax_nonPITsalesgen_nom_state = tax_tot_nom_state - tax_indivIncome_nom_state - tax_sales_gen_nom_state,
				 tax_nonPITsalestot_nom_state = tax_tot_nom_state - tax_indivIncome_nom_state - tax_sales_tot_nom_state,
				 NGSPc = ifelse(year > 1997, NGSP_NAICS, NGSP_SIC),
				 NGSPc = 1000 * NGSPc,
				 PIT_tax = 100 * tax_indivIncome_nom_state/tax_tot_nom_state, 
				 tax_GSP = 100 * tax_tot_nom_state / NGSPc)

df_dlnom <- 
	df_nom %>% 
	group_by(state_abb) %>% 
	mutate_at(vars(-year), funs(100 * log(./lag(.)))) %>% 
	mutate(NGSP = ifelse(year > 1997, NGSP_NAICS, NGSP_SIC))

df_nom
df_dlnom





#**********************************************************************
#       Descriptive analysis 1: structure of state tax revenue     ####
#**********************************************************************

# For US aggregate:
 # The share of sales tax is quite stable, ~total sales 50%, general sales tax 30%. 
 # The share of personal income taxes has increased by half, from 25% in 1977 to 37% in 2015 
 # The share of property tax is around 1.5%~2.5%. 
 # The share of corp income tax decreased from 9% to 5%

## Sales+PIT as a % of total tax revenue.
 # Total sales
 # tot sales + PIT accounts for 70%+ of tax revenue (most above 80%), 
 # States with sales + PIT less than 70% of tax rev are: AK(30%), WY(42%), NH(43%), ND(43%), DE(47%), VT(57%), MT(63%)

 # General sales
 # general sales + PIT: mostly 60%+ (68% for US)
 # States with general sales + PIT less than 50%:  AK, WY, NH, ND, DE, VT, MT


# Year 2015: ordered by share of individual income tax
 # Top Five in 2015:       OR(69%), VA(58%), NY(56%), MA(54%), CA(52%)
 # states with % PIT < 5%: NH, TN, AK(0), FL, NV, SD, TX WA, WY 


# Year 2015: ordered by share of sales tax
 # sales tot
 # Top five in 2015: TX(86.5%), SD(82%), FL(82%), NV(80%), WA(79%)
 # Lowest five in 2015: OR(14%), DE(14%), MT(21%), AK(30%), NY(31%)   [NH 39%, why]

 # sales general
 # Top five in 2015: TX(61%), WA(61%), FL(59%), SD(58%), NV(54%)
 # Lowest five in 2015(excluding five states with no gen sales): VT(12%), NY(17%), VA(18%), MA(21%), CO(22%)   

 # Five states do not have statewide sales taxes: Alaska, Delaware, Montana, New Hampshire, and Oregon. 
 # Of these, Alaska and Montana allow localities to charge local sales taxes.



## Structure of state tax revenue 1977-2015, census data (from Urban) based on nominal terms
taxStr_state_nom <- 
df_RevGSP %>% 
	filter(year %in% 1977:2015, state_abb %in% df_us_states$state_abb) %>% 
	select(state_abb, year, 
				 tax_tot_nom_state, 
				 tax_indivIncome_nom_state,
				 tax_sales_gen_nom_state,
				 tax_sales_tot_nom_state,
         tax_property_nom_state,
         tax_corpIncome_nom_state) %>% 
	mutate(indivIncome_pct = 100 * tax_indivIncome_nom_state/tax_tot_nom_state,
				 salesgen_pct    = 100 * tax_sales_gen_nom_state/tax_tot_nom_state,
				 salestot_pct    = 100 * tax_sales_tot_nom_state/tax_tot_nom_state,
				 property_pct    = 100 * tax_property_nom_state/tax_tot_nom_state,
				 corpIncome_pct  = 100 * tax_corpIncome_nom_state/tax_tot_nom_state,
				 PITsalesgen_pct = indivIncome_pct + salesgen_pct,
				 PITsalestot_pct = indivIncome_pct + salestot_pct,
				 nonPITsalesgen_pct  = 100 - PITsalesgen_pct,
				 nonPITsalestot_pct  = 100 - PITsalestot_pct
				 ) %>% 
  select(state_abb, year,
  			 indivIncome_pct,
  			 salesgen_pct,
  			 salestot_pct,
  			 property_pct,
  			 corpIncome_pct,
  			 PITsalesgen_pct,
  			 nonPITsalesgen_pct,
  			 PITsalestot_pct,
  			 nonPITsalestot_pct)

# Share of income tax
taxStr_state_nom %>% 
	filter(year == 2015) %>% 
	select(state_abb, year, indivIncome_pct, salesgen_pct, PITsalesgen_pct, nonPITsalesgen_pct) %>% 
	arrange(desc(indivIncome_pct))

# Share of gen sales tax
taxStr_state_nom %>% 
	filter(year == 2015) %>% 
	select(state_abb, year, indivIncome_pct, salesgen_pct, PITsalesgen_pct, nonPITsalesgen_pct) %>% 
	arrange(desc(salesgen_pct))


## Plotting % of sales against % of PIT
taxStr_state_nom %>% 
	filter(year == 2015, !is.na(state_abb)) %>% 
	ggplot(aes(x = indivIncome_pct, y = salesgen_pct, label = state_abb)) + 
	geom_point() +
	geom_text(nudge_x = 1, size = 2)



#*******************************************************************************
#   Descriptive analysis 2:  GSP growth and tax growth                      ####
#*******************************************************************************

## Key points to make
  # 1. For the US aggregate, income tax is more responsive to economic conditions than sales tax, 
  #    which is especially true during the last two recessions. 
  # 2. At state level, we have the same observation. There are variations in the PIT-GSP and sales_GSP relationship across states. 
  # 3. The relationship between PIT and GSP may not be linear. Since the share of capital gains and losses have been increasing 
  #    in income tax bases, it may make sense to add financial market variables as a determinants PIT. 
  #    This will also solve the issue in modeling PIT and GSP with a linear relationship that their growth rates must have the same sign.    
  # 4. Is sales tax linear to GSP? 
  # 5. About Non-PIT-non_salesgen taxes
  #     - select sales tax (about 50% of general sales tax) seems to be more stable than general 
  #     - corp income tax is the most volatile component, but its share is small: ~5%
  #     - The share of property tax is very small ~1.5%
  #     - Other taxes (8~10%), show some cyclicity
  #     - For now, may assume a fixed share of non_PIT_sales, say 15% in all taxes. Need to further check how to treat select sales tax 

# Look at total tax revenue of 5 states with the highest and the lowest share of PIT. 

states_PIT    <- c("US", "OR", "NY", "VA", "MA", "CA")
states_sales  <- c("US", "TX", "WA", "FL", "SD", "NV") # sales tax states with zero PIT
states_sales2 <- c("US", "HI", "AZ", "MS", "OH", "IN") # sales tax states with non-zero PIT


## 1. GSP and TOTAL tax revenue over time
 # US + PIT states
df_dlreal %>% 
	select(state_abb, year, RGSP, tax_tot_real_state) %>% 
	filter(year >= 1987, state_abb %in% states_PIT) %>% 
	gather(var, value, -state_abb, -year) %>% 
	ungroup() %>% 
	mutate(state_abb = factor(state_abb, levels = states_PIT)) %>% 
	ggplot(aes(x = year, y = value, color = var ))+ theme_bw() + facet_grid(state_abb~.) + 
	geom_line() +
	geom_point() + 
	geom_hline(yintercept = 1, linetype = 2) + 
	scale_color_manual(values = c("darkgrey", "blue"))+
	scale_x_continuous(breaks = seq(1950, 2015, 5)) + 
	coord_cartesian(ylim = c(-15, 15)) + 
	theme(legend.position="bottom") +
	labs(title = "PIT states")
	

 # US + sales states with no PIT
df_dlreal %>% 
	select(state_abb, year, RGSP, tax_tot_real_state) %>% 
	filter(year >= 1987, state_abb %in% states_sales) %>% 
	gather(var, value, -state_abb, -year) %>% 
	ungroup() %>% 
	mutate(state_abb = factor(state_abb, levels = states_sales)) %>% 
	ggplot(aes(x = year, y = value, color = var ))+ theme_bw() + facet_grid(state_abb~.) + 
	geom_line() +
	geom_point() + 
	geom_hline(yintercept = 1, linetype = 2) + 
	scale_color_manual(values = c("darkgrey", "blue"))+
	scale_x_continuous(breaks = seq(1950, 2015, 5)) + 
	coord_cartesian(ylim = c(-15, 15)) + 
	theme(legend.position="bottom") + 
	labs(title = "sales states")

 # US + sales states with PIT
df_dlreal %>% 
	select(state_abb, year, RGSP, tax_tot_real_state) %>% 
	filter(year >= 1987, state_abb %in% states_sales2) %>% 
	gather(var, value, -state_abb, -year) %>% 
	ungroup() %>% 
	mutate(state_abb = factor(state_abb, levels = states_sales2)) %>% 
	ggplot(aes(x = year, y = value, color = var ))+ theme_bw() + facet_grid(state_abb~.) + 
	geom_line() +
	geom_point() + 
	geom_hline(yintercept = 1, linetype = 2) + 
	scale_color_manual(values = c("darkgrey", "blue"))+
	scale_x_continuous(breaks = seq(1950, 2015, 5)) + 
	coord_cartesian(ylim = c(-15, 15)) + 
	theme(legend.position="bottom") + 
	labs(title = "sales states")


## 2.  GSP and income tax and sales tax
 # Top 5 income tax states
df_dlreal %>% 
	select(state_abb, year, RGSP, tax_indivIncome_real_state, tax_sales_gen_real_state) %>% 
	filter(year >= 1987, state_abb %in% states_PIT) %>% 
	gather(var, value, -state_abb, -year) %>% 
	ungroup() %>% 
	mutate(state_abb = factor(state_abb, levels = states_PIT)) %>% 
	ggplot(aes(x = year, y = value, color = var ))+ theme_bw() + facet_grid(state_abb~.) + 
	geom_line() +
	geom_point() + 
	geom_hline(yintercept = 1, linetype = 2) + 
	scale_color_manual(values = c("darkgrey", "blue", "lightblue"))+
	scale_x_continuous(breaks = seq(1950, 2015, 5)) + 
	coord_cartesian(ylim = c(-20, 20)) + 
	theme(legend.position="bottom") +
	labs(title = "PIT")


 # Top 5 sales tax states with no PIT
df_dlreal %>% 
	select(state_abb, year, RGSP, tax_indivIncome_real_state, tax_sales_gen_real_state) %>% 
	filter(year >= 1987, state_abb %in% states_sales) %>% 
	gather(var, value, -state_abb, -year) %>% 
	ungroup() %>% 
	mutate(state_abb = factor(state_abb, levels = states_sales)) %>% 
	ggplot(aes(x = year, y = value, color = var ))+ theme_bw() + facet_grid(state_abb~.) + 
	geom_line() +
	geom_point() + 
	geom_hline(yintercept = 1, linetype = 2) + 
	scale_color_manual(values = c("darkgrey", "blue", "lightblue"))+
	scale_x_continuous(breaks = seq(1950, 2015, 5)) + 
	coord_cartesian(ylim = c(-15, 15)) + 
	theme(legend.position="bottom") +
	labs(title = "sales1")

# 5 sales tax states with PIT
df_dlreal %>% 
	select(state_abb, year, RGSP, tax_indivIncome_real_state, tax_sales_gen_real_state) %>% 
	filter(year >= 1987, state_abb %in% states_sales2) %>% 
	gather(var, value, -state_abb, -year) %>% 
	ungroup() %>% 
	mutate(state_abb = factor(state_abb, levels = states_sales2)) %>% 
	ggplot(aes(x = year, y = value, color = var ))+ theme_bw() + facet_grid(state_abb~.) + 
	geom_line() +
	geom_point() + 
	geom_hline(yintercept = 1, linetype = 2) + 
	scale_color_manual(values = c("darkgrey", "blue", "lightblue"))+
	scale_x_continuous(breaks = seq(1950, 2015, 5)) + 
	coord_cartesian(ylim = c(-15, 15)) + 
	theme(legend.position="bottom") +
	labs(title = "sales2")


## 3. non-PIT-non-sales 

# Top 5 income tax states
df_dlreal %>% 
	select(state_abb, year, RGSP, tax_indivIncome_real_state, tax_sales_gen_real_state, tax_nonPITsalestot_real_state) %>% 
	filter(year >= 1987, state_abb %in% states_PIT) %>% 
	gather(var, value, -state_abb, -year) %>% 
	ungroup() %>% 
	mutate(state_abb = factor(state_abb, levels = states_PIT)) %>% 
	ggplot(aes(x = year, y = value, color = var ))+ theme_bw() + facet_grid(state_abb~.) + 
	geom_line() +
	geom_point() + 
	geom_hline(yintercept = 1, linetype = 2) + 
	scale_color_manual(values = c("darkgrey", "blue", "lightblue", "lightgreen"))+
	scale_x_continuous(breaks = seq(1950, 2015, 5)) + 
	coord_cartesian(ylim = c(-20, 20)) + 
	theme(legend.position="bottom") +
	labs(title = "PIT")

# 
fct_order = c( "tax_tot_real_state",
	             "tax_indivIncome_real_state",
							 "tax_sales_gen_real_state",
							 "tax_sales_select_real_state",
							 "tax_corpIncome_real_state",
							 "tax_property_real_state",
							 "tax_other_real_state")

fct_label = c("Total tax",
	            "PIT",
							"sales_general",
							"sales_select",
							"corpIncome",
							"property",
							"other")

df <- 
df_dlreal %>% 
	filter(year %in% 1988:2015, state_abb %in% df_us_states$state_abb) %>% 
	select(state_abb, year, 
				 RGSP,
				 tax_tot_real_state,
				 tax_indivIncome_real_state,
				 tax_sales_gen_real_state,
				 tax_sales_select_real_state,
				 tax_corpIncome_real_state,
				 tax_property_real_state,
				 tax_other_real_state) %>% 
	gather(taxtype, value, -state_abb, -year, -RGSP) %>% 
	mutate(var = "tax")

df1 <- df %>% select(state_abb, year, taxtype, var, value)
df2 <- df %>% mutate(value = RGSP, var = "RGSP") %>% select(state_abb, year, taxtype, var, value) 
df3 <- bind_rows(df1, df2) %>% 
	mutate(taxtype = factor(taxtype, levels = fct_order, labels = fct_label))



df3 %>% filter(state_abb == "US") %>% 
	ggplot(aes(x = year, y = value, color = var)) + theme_bw() + 
	facet_grid(taxtype~.) + 
	geom_line() + 
	geom_point() + 
	geom_hline(yintercept = 1, linetype = 2) +
	scale_color_manual(values = c("darkgrey", "blue"))+
	coord_cartesian(ylim = c(-25, 25))
	
df_real_pct <- 
df_real %>% 
	filter(state_abb %in% df_us_states$state_abb) %>% 
	mutate(
				 pct_indivIncome  = 100 * tax_indivIncome_real_state/tax_tot_real_state,
				 pct_sales_gen    = 100 * tax_sales_gen_real_state/tax_tot_real_state,
				 pct_sales_select = 100 * tax_sales_select_real_state/tax_tot_real_state,
				 pct_corpIncome   = 100 * tax_corpIncome_real_state/tax_tot_real_state,
				 pct_property     = 100 * tax_property_real_state/tax_tot_real_state,
				 pct_other        = 100 * tax_other_real_state/tax_tot_real_state ) %>% 
	select(state_abb, year, 
				 pct_indivIncome, 
				 pct_sales_gen,   
				 pct_sales_select,
				 pct_corpIncome,  
				 pct_property,   
				 pct_other) 
	

df_real_pct %>%
	filter(state_abb == "US")
	
df_real_pct %>%
	filter(year == 2015, state_abb %in% states_PIT)

df_real_pct %>%
	filter(year == 2015, state_abb %in% states_sales)

df_real_pct %>%
	filter(year == 2015, state_abb %in% states_sales2)



#*******************************************************************************
#                 Regression analysis: data preparation 1        ####
#*******************************************************************************


# GSP and real tax revenue data
# Notes:
  #  GSP from 1988-2016
  #  tax from 1978-2015
  #  Common period:1988-2015
  #  Nominal GSP goes back to 1964, may want to extend the real GSP series by adjusting pre-1988 nominal GSP for inflation   

df_dlRevGSP_reg <- 
df_dlreal %>% 
	select(state_abb, year,
				 RGSP,
				 tax_tot_real_state, 
				 tax_indivIncome_real_state,
				 tax_sales_gen_real_state,
				 tax_sales_select_real_state,
				 tax_sales_tot_real_state,
				 tax_corpIncome_real_state,
				 tax_property_real_state,
				 tax_other_real_state,
				 tax_nonPITsalestot_real_state) %>% 
	filter(year %in% 1978:2015, state_abb %in% df_us_states$state_abb) %>% 
	mutate_at(vars(-year), funs(./100))
df_dlRevGSP_reg 







# Asset return data from SBBI and national GDP (FRED, check it against the aggregate GSP for all states)
df_dataAll_y %>% names()

df_SBBI_reg <- 
df_dataAll_y %>% 
	select(year, month, yearMon, 
				 GDP_FRED,        # National GDP from FRED
				 CPIU_NA_FRED,    # CPI, urban resident, not seasonally adjusted. FRED
				 Inflation_Index, # SBBI, inflation index 
				 LCapStock_TRI,   # SBBI, large cap stock total return index (SP500 total return index)
				 LCapStock_CAI,   # SBBI, large cap stock capital appreciation index (SP500 price index)
				 CBond_TRI,       # SBBI, corp bond total return
				 LTGBond_TRI,     # SBBI, long-term gov bond total return (20y?)
				 MTGBond_TRI,     # SBBI, medium-term gov bond total return
				 LTGBond_Yield    # SBBI, long-term gov bond yield
				 ) %>%  
  filter(year %in% 1977:2015) %>% 
	mutate_at(vars(-year, -month, -yearMon, -LTGBond_Yield), funs(log(./lag(.))) )
df_SBBI_reg 



# Asset return data from NYU Stern 
df_assetReturn_nyu <- read_excel(paste0(dir_data_raw, "/histretSP_201801.xls"), sheet = "Returns by year", skip = 17)[,1:4]
names(df_assetReturn_nyu) <- c("year", "SP500_return_nyu", "TBill_return_nyu", "TBond_return_nyu")
df_assetReturn_nyu %<>% mutate(year = as.numeric(year)) %>% filter(year <= 2017)
df_assetReturn_nyu


# Captial gain / losses data for states
  # From Don
  # level in $billions
rlzCapGains_nom <- read_excel(paste0(dir_data_raw, "/NationalCapitalGains_yy.xlsx"), sheet = "CapGains", range = c("A5:C67"))
names(rlzCapGains_nom) <- c("year", "capgains", "capgains_chg")
rlzCapGains_nom

rlzCapGains_nom_rev <- read_excel(paste0(dir_data_raw, "/NationalCapitalGains_yy.xlsx"), sheet = "CapGains_rev", range = c("A8:E30"))[,-(2:3)]
rlzCapGains_nom_rev


df_real_reg <- 
	df_dlRevGSP_reg %>% 
	left_join(df_SBBI_reg) %>% 
	left_join(df_assetReturn_nyu) %>% 
	left_join(rlzCapGains_nom) %>% 
	left_join(rlzCapGains_nom_rev)

df_real_reg # all variables 1988-2015


## Extending RGSP series by adjusting NGSP for inflation 

df_GSPext <- df_dlnom %>% select(state_abb, year, NGSP) %>% 
	           left_join(df_SBBI_reg %>% select(year, Inflation_Index)) %>% 
	           mutate(RGSP_ext = NGSP/100 - Inflation_Index) %>% 
	           select(state_abb, year, RGSP_ext)
     
df_real_reg %<>% left_join(df_GSPext) %>% 
	mutate(RGSP_ext = ifelse(year < 1988, RGSP_ext, RGSP ))


df_real_reg %>% 
	select(state_abb, year, RGSP_ext, RGSP) %>% 
	filter(state_abb == "US") %>% 
	gather(var, value, -state_abb, -year) %>% 
	qplot(x = year, y = value, color = var, data =.,  geom = c("line", "point"))






#*******************************************************************************
#     Regression analysis: data prep 2 asset return and PIT                 ####
#*******************************************************************************

# Goal: Estimate the relationship between real PIT growth, real GSP growth, and asset return (nominal),
#       and see if there is structure breek around 2000. (if the recent two recessions are different.)

# Notes:
 # Need to first examine the relationship between realized capgains, tax revenue from realized capgains, and asset returns.   
 # Need an argument for using investment returns in the PIT regression




# 1. Asset returns and captial gains, realized captial gains and PIT

df_temp <- 
df_real_reg %>% 
	select(state_abb, year, 
				 LCapStock_CAI, LCapStock_TRI, 
				 LTGBond_TRI,   LTGBond_Yield, 
				 SP500_return_nyu, 
				 TBond_return_nyu, 
				 capgains_chg, 
				 tax_gains_chg, 
				 tax_indivIncome_real_state) %>% 
	filter(state_abb == "US")

# quick look at how the total bond returns from SBBI and NYU Stern differ
df_temp %>% 
	select(state_abb, year, LTGBond_TRI, TBond_return_nyu) %>% 
	gather(var, value, -state_abb, -year) %>% 
	qplot(x = year, y = value, color = var, data=.,  geom = c("line", "point")) + theme_bw() + 
	scale_x_continuous(breaks = seq(1950, 2020, 5))
# The two sources show very different returns, especially after 2000. 
# We may want to stick with the SBBI return. (SBBI's calculation looks more rigorous) 


df_temp %>% 
	select(state_abb, year, LCapStock_TRI, SP500_return_nyu) %>% 
	gather(var, value, -state_abb, -year) %>% 
	qplot(x = year, y = value, color = var, data=.,  geom = c("line", "point")) + theme_bw() + 
	scale_x_continuous(breaks = seq(1950, 2020, 5))
# It seems there is a lag in SBBI SP500 return


# Plotting asset returns and realized capital gains 
# Stock
fig_returnCapGains <- 
df_temp %>% 
	select(state_abb, year, LCapStock_TRI, capgains_chg) %>%  
	gather(var, value, -state_abb, -year) %>% 
	mutate(var = factor(var, levels = c("LCapStock_TRI", "capgains_chg"),
											     labels = c("SP500 Total Return", "Change in realized capital gains"))) %>% 
	qplot(x = year, y = 100 * value, color = var, data=.,  geom = c("line", "point")) + 
	theme_bw() + RIG.themeLite() +  
	
	scale_x_continuous(breaks = seq(1950, 2020, 5)) + 
	scale_color_manual(values = c(RIG.green, "blue")) + 
	labs(x = NULL, y = "Percent", color = NULL,
			 title = "Total stock returns and realized capital gains") + 
	theme(legend.position = "bottom")
fig_returnCapGains


# Long-term gov bond, SBBI
df_temp %>% 
	select(state_abb, year, LTGBond_TRI, capgains_chg) %>%  
	gather(var, value, -state_abb, -year) %>% 
	qplot(x = year, y = value, color = var, data=.,  geom = c("line", "point")) + theme_bw() + 
	scale_x_continuous(breaks = seq(1950, 2020, 5))
# No obvious correlation 

# Long-term gov bond, NYU
df_temp %>% 
	select(state_abb, year, TBond_return_nyu, capgains_chg) %>%  
	gather(var, value, -state_abb, -year) %>% 
	qplot(x = year, y = value, color = var, data=.,  geom = c("line", "point")) + theme_bw() + 
	scale_x_continuous(breaks = seq(1950, 2020, 5))
#


# Plotting realized captial gains 
 # High correlation between 1-year lag of realized captial gain and capital gains tax receipts 
 # High correlation between captial gain tax receipt and state PIT
 # Question: the share of capital gain tax in total PIT

fig_captainsTax <- 
df_temp	%>% 
	select(state_abb, year, capgains_chg, tax_gains_chg, tax_indivIncome_real_state) %>% 
	mutate(capgains_chg = lag(capgains_chg)) %>% 
	gather(var, value, -state_abb, -year) %>% 
	mutate(var = factor(var, levels = c("capgains_chg", "tax_gains_chg", "tax_indivIncome_real_state"),
											     labels = c("Change in realized capital gains (1 year lag)", "Change in capital gains tax receipts", "change in state individual tax"))) %>%
	qplot(x = year, y = 100 * value, color = var, data=.,  geom = c("line", "point")) + 
	theme_bw() + RIG.themeLite() + 
	geom_hline(yintercept = 0, linetype = 2) + 
	scale_x_continuous(breaks = seq(1950, 2020, 5)) + 
	scale_color_manual(values = c("blue", RIG.red, "darkgrey")) + 
	labs(x = NULL, y = "Percent", color = NULL,
			 title = "Realized capital gains and tax revenues") + 
	theme(legend.position = "bottom")
fig_captainsTax




#*******************************************************************************
#     Regression analysis US: PIT               ####
#*******************************************************************************

# National level regressions
df_reg_temp <- 
	df_real_reg %>% 
	select(state_abb, year, 
				 tax_indivIncome_real_state, 
				 tax_sales_gen_real_state,
				 tax_sales_select_real_state,
				 tax_sales_tot_real_state,
				 tax_nonPITsalestot_real_state,
				 RGSP,
				 RGSP_ext,
				 GDP_FRED,
				 LCapStock_TRI,
				 LTGBond_TRI,
				 capgains_chg,
				 tax_gains_chg) %>% 
	mutate(Lagcapgains_chg = lag(capgains_chg),
				 after99 = ifelse(year > 2001, TRUE, FALSE),
				 recession = ifelse(year %in% c(2001:2002, 2008:2009), TRUE, FALSE),
				 RGSP_after99 = RGSP*after99)


# PIT, National level

# GSP only
df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%
	lm(tax_indivIncome_real_state ~ RGSP, data = .) %>% 
	summary

df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%
	lm(tax_indivIncome_real_state ~ RGSP + RGSP:after99, data = .) %>% 
	summary

df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%
	lm(tax_indivIncome_real_state ~ RGSP:after99, data = .) %>% 
	summary



# GSP and SP500 total return
df_reg_temp %>% filter(state_abb == "US") %>% 
	#mutate(RGSP = RGSP_ext) %>% 
	mutate(RGSP = GDP_FRED) %>% 
	lm(tax_indivIncome_real_state ~ RGSP + LCapStock_TRI, data = .) %>% 
	summary
  # only SP500 is significant

df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>% 
	#mutate(RGSP = RGSP_ext) %>% 
	lm(tax_indivIncome_real_state ~ RGSP + RGSP:after99 + LCapStock_TRI + LCapStock_TRI:after99, data =.) %>% 
	summary

df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>% 
	#mutate(RGSP = RGSP_ext) %>% 
	lm(tax_indivIncome_real_state ~ RGSP + LCapStock_TRI + LCapStock_TRI:after99, data =.) %>% 
	summary
# No variable is significant even at 10% level
# but passes F test. 



df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>% 
	lm(tax_indivIncome_real_state ~ RGSP:after99 + LCapStock_TRI:after99, data =.) %>% 
	summary
df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>% 
	lm(tax_indivIncome_real_state ~ RGSP + LCapStock_TRI:after99, data =.) %>% 
	summary
  # This is similar to run regressions for data before and after 1999 separately 
  # Only SP500 after 1999 is significant





# GSP and lag of capital gains
df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%
	lm(tax_indivIncome_real_state ~ RGSP + Lagcapgains_chg, data = .) %>% 
	summary
  # Extremely significant. very high R-squared. RGSP is insignificant (p = 0.13)


df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%
	lm(tax_indivIncome_real_state ~ RGSP + RGSP:after99 + Lagcapgains_chg+  Lagcapgains_chg:after99, data = .) %>% 
	summary
# GSP before 1999 and SP500 after 1999 (10% level) are significant
df_reg_temp %>% filter(state_abb == "US") %>% 
	lm(tax_indivIncome_real_state ~ RGSP + Lagcapgains_chg + Lagcapgains_chg:after99, data = .) %>% 
	summary
# GSP and SP500 after 1999 (10% level) are significant


df_reg_temp %>% filter(state_abb == "US") %>% 
	lm(tax_indivIncome_real_state ~ RGSP:after99 +  Lagcapgains_chg:after99, data = .) %>% 
	summary
df_reg_temp %>% filter(state_abb == "US") %>% 
	lm(tax_indivIncome_real_state ~ RGSP + Lagcapgains_chg:after99, data = .) %>% 
	summary







# GSP and capital gains taxes (from 1996) 
df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%
	lm(tax_indivIncome_real_state ~ RGSP + tax_gains_chg, data = .) %>% 
	summary

df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%
	lm(tax_indivIncome_real_state ~ RGSP:after99 + tax_gains_chg:after99, data = .) %>% 
	summary

df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%
	lm(tax_indivIncome_real_state ~ RGSP + tax_gains_chg:after99, data = .) %>% 
	summary


# Extremely significant, of course. similar R-squared as capital gains


## Using Stock returns as a proxy for capital gains losses. 
df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%
	lm(capgains_chg ~ LCapStock_TRI, data = .) %>% 
	summary
  # intercept not significant, parameter close to 1. 


df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%
	lm(capgains_chg ~ LCapStock_TRI + LCapStock_TRI:after99, data = .) %>% 
	summary


df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%
	lm(capgains_chg ~ LCapStock_TRI:after99, data = .) %>% 
	summary
  # if regressed separately, response to stock return is higher after 1999.   


#*******************************************************************************
#     Regression analysis US: Sales                ####
#*******************************************************************************

# General sales National level
df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%
	lm(tax_sales_gen_real_state ~ RGSP, data = .) %>% 
	summary

df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%
	lm(tax_sales_gen_real_state ~ RGSP + RGSP:recession, data = .) %>%  # with recession dummy
	summary

df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%                                         # with break dummy
	lm(tax_sales_gen_real_state ~ RGSP + RGSP:after99, data = .) %>% 
	summary

# elasticity is about 1.2 (note this is smaller the GSP elasticity of PIT when SP500 is present) 
# GSP/after1999 interaction not significant at 10%, but p is 0.12
# GSP/recession interaction is NOT significant


# Select sales
df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%  
	lm(tax_sales_select_real_state ~ RGSP, data = .) %>% 
	summary

df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%  
	lm(tax_sales_select_real_state ~ RGSP + RGSP:after99, data = .) %>% 
	summary

df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%  
	lm(tax_sales_select_real_state ~ RGSP + RGSP:recession, data = .) %>% 
	summary

# select sales tax: elasticity is about 0.5. 
# 

# Select sales
df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%  
	lm(tax_sales_tot_real_state ~ RGSP, data = .) %>% 
	summary

df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%  
	lm(tax_sales_tot_real_state ~ RGSP + RGSP:after99, data = .) %>% 
	summary

df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%  
	lm(tax_sales_tot_real_state ~ RGSP + RGSP:recession, data = .) %>% 
	summary

# elasticity for sales tax is about 0.95~1. 
# This is consistent what we can infer from the separate model results of general and select sales tax and their share in total sales




#*******************************************************************************
#     Regression analysis US: Other             ####
#*******************************************************************************

# non PIT/sales National level (corp income + property + other)

df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>% 
	lm(tax_nonPITsalestot_real_state ~ RGSP, data = .) %>% 
	summary
# significant, largely attributable to corp income?


df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%
	lm(tax_nonPITsalestot_real_state ~ RGSP + RGSP:after99, data = .) %>% 
	summary
# interaction with after1999 significant


df_reg_temp %>% filter(state_abb == "US") %>% 
	mutate(RGSP = GDP_FRED) %>%
	lm(tax_nonPITsalestot_real_state ~ RGSP + RGSP:recession, data = .) %>% 
	summary
# interaction with recession not significant


#*******************************************************************************
#       Regression analysis panel: PIT               ####
#*******************************************************************************
library(plm)


# Continent states only, excluding states with no PIT
df_reg_plm_PIT <-  df_reg_temp %>% 
	filter(!state_abb %in% c("US", "DC", "AL", "HI", "AK", "FL", "NV", "SD", "TX", "WA", "WY")) %>% 
	pdata.frame(index("state_abb", "year"))

# df_reg_plm_PIT %>% filter(is.na(RGSP_ext), year == 2013)


# No breaking point at 1999
df_reg_plm_PIT %>% 
	filter(year %in% 1978:2015) %>% 
	plm(tax_indivIncome_real_state ~ RGSP_ext, data =., model = "within") %>% 
	summary
	
df_reg_plm_PIT %>% 
	filter(year %in% 1978:2015) %>% 
	plm(tax_indivIncome_real_state ~ RGSP_ext + LCapStock_TRI, data =., model = "within") %>% 
	summary


# breaking point at 1999
df_reg_plm_PIT %>%   # GSP only
	filter(year %in% 1978:2015) %>% 
	mutate(RGSP = RGSP_ext) %>% 
	plm(tax_indivIncome_real_state ~ RGSP + RGSP:after99, data =., model = "within") %>% 
	summary

df_reg_plm_PIT %>%   # break for GSP and sp500
	filter(year %in% 1978:2015) %>% 
	mutate(RGSP = RGSP_ext) %>% 
	plm(tax_indivIncome_real_state ~ RGSP + RGSP:after99 + LCapStock_TRI + LCapStock_TRI:after99, data =., model = "within") %>% 
	summary

df_reg_plm_PIT %>%  # break for sp500 only 
	filter(year %in% 1988:2015) %>% 
	#mutate(RGSP = RGSP_ext) %>% 
	plm(tax_indivIncome_real_state ~ RGSP + LCapStock_TRI + LCapStock_TRI:after99, data =., model = "within") %>% 
	summary


# Estimation for states separately
df_reg_plm_PIT %>% 
	filter(year %in% 1978:2015) %>% 
	pvcm(tax_indivIncome_real_state ~ RGSP_ext + LCapStock_TRI + LCapStock_TRI:after99, data =., model = "within") 


# Pooling test
df_reg_plm_PIT %>% 
	filter(year %in% 1978:2015) %>% 
	mutate(RGSP = RGSP_ext) %>%
  pooltest(tax_indivIncome_real_state ~ RGSP + LCapStock_TRI + LCapStock_TRI:after99, data =., model = "within")
# Cannot reject the H0 (poolable)
# However, when estimated separately, the estimates vary greatly across states


# Hausman test
gw <- 
  df_reg_plm_PIT %>% 
	filter(year %in% 1978:2015) %>% 
	mutate(RGSP = RGSP_ext) %>%
	plm(tax_indivIncome_real_state ~ RGSP_ext + LCapStock_TRI + LCapStock_TRI:after99, data =., model = "within")

gr <- 
	df_reg_plm_PIT %>% 
	filter(year %in% 1978:2015) %>% 
	mutate(RGSP = RGSP_ext) %>%
	plm(tax_indivIncome_real_state ~ RGSP_ext + LCapStock_TRI + LCapStock_TRI:after99, data =., model = "random")

phtest(gw, gr) # cannot reject H0 that both are consistent (fixed effect prefered.




#*******************************************************************************
#       Regression analysis panel: sales             ####
#*******************************************************************************
# Notes
 # Poolability rejected, FE preferred
 # dummary after 1999 insignificant
 # Elasticity estimated by FE is around 0.77. This is even slightly higher than the FE estimate of the elasticity of PIT
 # Elasticity of select sales tax (FE) is 0.45

# Continent states only, excluding states with no general sales
df_reg_plm_sales <-  df_reg_temp %>% 
	filter(!state_abb %in% c("US", "DC", "AL", "AK", "HI", "DE", "MT", "NH", "OR")) %>% 
	pdata.frame(index("state_abb", "year"))

#df_reg_plm_sales_gen %>% filter(is.na(tax_sales_gen_real_state), year == 2013)


# No breaking point at 1999
df_reg_plm_sales_gen %>% 
	filter(year %in% 1978:2015) %>% 
	plm(tax_sales_gen_real_state ~ RGSP_ext, data =., model = "within") %>% 
	summary

df_reg_plm_sales_gen %>% 
	filter(year %in% 1978:2015) %>% 
	plm(tax_sales_gen_real_state ~ RGSP_ext + RGSP_ext:after99, data =., model = "within") %>% 
	summary 
  # dummy after99 not significant

df_reg_plm_sales_gen %>% 
	filter(year %in% 1978:2015) %>% 
	plm(tax_sales_select_real_state ~ RGSP_ext, data =., model = "within") %>% 
	summary




# Estimation for states separately
df_reg_plm_sales %>% 
	filter(year %in% 1978:2015) %>% 
	pvcm(tax_sales_gen_real_state ~ RGSP_ext , data =., model = "within") %>% 
	summary


# Pooling test
df_reg_plm_sales %>% 
	filter(year %in% 1978:2015) %>% 
	pooltest(tax_sales_gen_real_state~ RGSP_ext, data =., model = "within")
# Poolability is rejected



# Hausman test
gw <- 
	df_reg_plm_sales %>% 
	filter(year %in% 1978:2015) %>% 
	mutate(RGSP = RGSP_ext) %>%
	plm(tax_sales_gen_real_state ~ RGSP_ext, data =., model = "within")

gr <- 
	df_reg_plm_sales %>% 
	filter(year %in% 1978:2015) %>% 
	mutate(RGSP = RGSP_ext) %>%
	plm(tax_sales_gen_real_state ~ RGSP_ext, data =., model = "random")

phtest(gw, gr) # cannot reject H0 that both are consistent at 5% level (fixed effect prefered).
               # but p-value is only 0.05


#*******************************************************************************
#       Regression analysis panel: other             ####
#*******************************************************************************
# Notes
# Poolability rejected, FE preferred
# dummary after 1999 insignificant
# Elasticity estimated by FE is around 0.77. This is even slightly higher than the FE estimate of the elasticity of PIT
# Elasticity of select sales tax (FE) is 0.45

# Continent states only, excluding states with no general sales
df_reg_plm_other <-  df_reg_temp %>% 
	filter(!state_abb %in% c("US", "DC")) %>% 
	pdata.frame(index("state_abb", "year"))

# df_reg_plm_other %>% filter(is.na(tax_nonPITsalestot_real_state), year == 2013)


# No breaking point at 1999
df_reg_plm_other %>% 
	filter(year %in% 1978:2015) %>% 
	plm(tax_nonPITsalestot_real_state ~ RGSP_ext, data =., model = "within") %>% 
	summary # 0.91

df_reg_plm_other %>% 
	filter(year %in% 1978:2015) %>% 
	plm(tax_nonPITsalestot_real_state ~ RGSP_ext + RGSP_ext:after99, data =., model = "within") %>% 
	summary 
# dummy after99 significant, would double elasticity 




# Estimation for states separately
df_reg_plm_other %>% 
	filter(year %in% 1978:2015) %>% 
	pvcm(tax_nonPITsalestot_real_state ~ RGSP_ext , data =., model = "within") %>% 
	summary


# Pooling test
df_reg_plm_other %>% 
	filter(year %in% 1978:2015) %>% 
	pooltest(tax_nonPITsalestot_real_state~ RGSP_ext, data =., model = "within")
# Poolability cannot be rejected



# Hausman test
gw <- 
	df_reg_plm_other %>% 
	filter(year %in% 1978:2015) %>% 
	mutate(RGSP = RGSP_ext) %>%
	plm(tax_nonPITsalestot_real_state ~ RGSP_ext, data =., model = "within")

gr <- 
	df_reg_plm_other %>% 
	filter(year %in% 1978:2015) %>% 
	mutate(RGSP = RGSP_ext) %>%
	plm(tax_nonPITsalestot_real_state ~ RGSP_ext, data =., model = "random")

phtest(gw, gr) # cannot reject H0 that both are consistent (fixed effect prefered).








#*******************************************************************************
#                                Saving outputs                             ####
#*******************************************************************************

ggsave(paste0(dir_fig_out, "fig_returnCapGains.png"), fig_returnCapGains, width = 10*0.8, height = 5*0.8)
ggsave(paste0(dir_fig_out, "fig_captainsTax.png"),    fig_captainsTax,    width = 10*0.8, height = 5*0.8)