Fix small sample correction handling for HAC

[s3alfisc]

Core things to remember:

• For G.adj, use the number of unique time units
• "nonnested" is not relevant

Some R code to replicate fixest behavior:

library(fixest)

set.seed(123)
n_id <- 100        
n_time <- 10      
id <- rep(1:n_id, each = n_time)
time <- rep(1:n_time, times = n_id)
x1 <- rnorm(n_id * n_time)
x2 <- runif(n_id * n_time, 0, 10)
individual_fe <- rnorm(n_id)
time_fe <- rnorm(n_time)
y <- 1 + 0.5 * x1 - 0.3 * x2 + individual_fe[id] + time_fe[time] + rnorm(n_id * n_time)
panel_data <- data.frame(id = id, time = time, x1 = x1, x2 = x2, y = y)
head(panel_data)

panel_unit1 = panel_data[panel_data$id == 1,]

fit = feols(y ~ x1 | id + time, data = panel_data)

fit1 = feols(y ~ x1, data = panel_unit1)

a = vcov_NW(fit, lag = 4, time = ~ time, unit = ~ id, ssc = ssc(K.adj = TRUE, G.adj = FALSE, K.fixef = "none"))
b = vcov_NW(fit, lag = 4, time = ~ time, unit = ~ id, ssc = ssc(K.adj = TRUE, G.adj = FALSE, K.fixef = "nonnested"))
c = vcov_NW(fit, lag = 4, time = ~ time, unit = ~ id, ssc = ssc(K.adj = TRUE, G.adj = FALSE, K.fixef = "full"))

a / b
#            (Intercept) x1
#(Intercept)           1  1
#x1                    1  1
a / c
#            (Intercept) x1
#(Intercept)           1  1
#x1                    1  1

b / c

Note: G.adjarg is applied, inconsistent with new hetero behavior. See here lrberge/fixest#611 (comment)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Flag	Coverage Δ
core-tests	76.32% <100.00%> (-0.05%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
pyfixest/estimation/feols_.py	85.28% <100.00%> (-0.24%)	⬇️
pyfixest/utils/utils.py	87.14% <100.00%> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.