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
use crate::{FixedEffects, GreenersError};
use ndarray::{s, Array1, Array2};
use std::fmt;
#[derive(Debug)]
pub struct ThresholdResult {
pub threshold_gamma: f64,
pub params_regime1: Array1<f64>, // Coeficientes quando q <= gamma
pub params_regime2: Array1<f64>, // Coeficientes quando q > gamma
pub r_squared: f64,
pub ssr_min: f64,
pub n_search: usize, // Quantos candidatos testamos
}
impl fmt::Display for ThresholdResult {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
writeln!(f, "\n{:=^78}", " Panel Threshold Model (Hansen 1999) ")?;
writeln!(
f,
"{:<25} {:>10.4}",
"Estimated Threshold (Gamma):", self.threshold_gamma
)?;
writeln!(
f,
"{:<25} {:>10.4}",
"R-squared (Combined):", self.r_squared
)?;
writeln!(f, "{:<25} {:>10.4e}", "Min SSR:", self.ssr_min)?;
writeln!(f, "\n{:-^78}", " Regime 1 (Below Threshold) ")?;
writeln!(f, "{:<10} | {:>12}", "Variable", "Coef")?;
for i in 0..self.params_regime1.len() {
writeln!(f, "x{:<9} | {:>12.4}", i, self.params_regime1[i])?;
}
writeln!(f, "\n{:-^78}", " Regime 2 (Above Threshold) ")?;
writeln!(f, "{:<10} | {:>12}", "Variable", "Coef")?;
for i in 0..self.params_regime2.len() {
writeln!(f, "x{:<9} | {:>12.4}", i, self.params_regime2[i])?;
}
writeln!(f, "{:=^78}", "")
}
}
pub struct PanelThreshold;
impl PanelThreshold {
/// Estima o modelo de limiar (Single Threshold).
/// Grid Search sobre 'q' para encontrar o ponto de quebra ótimo.
pub fn fit(
y: &Array1<f64>,
x: &Array2<f64>,
q: &Array1<f64>, // Variável de Limiar (Threshold Variable)
entity_ids: &Array1<i64>,
) -> Result<ThresholdResult, GreenersError> {
let n = y.len();
let k = x.ncols();
if q.len() != n || entity_ids.len() != n {
return Err(GreenersError::ShapeMismatch("Input lengths differ".into()));
}
// 1. Definir Grid de Busca (Trimmed)
// Precisamos dos valores únicos de q, ordenados
let mut q_vec: Vec<f64> = q.to_vec();
q_vec.sort_by(|a, b| a.partial_cmp(b).unwrap());
q_vec.dedup(); // Apenas únicos
let n_unique = q_vec.len();
// Hansen recomenda descartar 15% das pontas (trimming parameter)
// para evitar regimes com pouquíssimos dados.
let trim_idx = (n_unique as f64 * 0.15).ceil() as usize;
if n_unique < 2 * trim_idx + 5 {
return Err(GreenersError::OptimizationFailed); // "Not enough variability in threshold variable"
}
let candidates = &q_vec[trim_idx..(n_unique - trim_idx)];
// Otimização: Se houver muitos candidatos (>300), pular alguns para velocidade
let step = if candidates.len() > 300 {
candidates.len() / 100
} else {
1
};
let mut best_gamma = 0.0;
let mut min_ssr = f64::INFINITY;
let mut best_params = Array1::<f64>::zeros(2 * k); // Vai guardar [Beta1, Beta2]
let mut best_r2 = 0.0;
// IDs cru para o FE
let id_slice = entity_ids.as_slice().unwrap();
// 2. Loop de Grid Search
for i in (0..candidates.len()).step_by(step) {
let gamma = candidates[i];
// Construir Matriz Expandida [X_low, X_high]
// X_low = X * I(q <= gamma)
// X_high = X * I(q > gamma)
// É mais eficiente criar vetores flat e transformar em Array2
let mut x_expanded_vec = Vec::with_capacity(n * 2 * k);
for row_idx in 0..n {
let q_val = q[row_idx];
let x_row = x.row(row_idx);
// if q_val <= gamma {
// // Regime 1 Ativo: [x, 0]
// for val in x_row {
// x_expanded_vec.push(*val);
// }
// for _ in 0..k {
// x_expanded_vec.push(0.0);
// }
// // Regime 2 Ativo: [0, x]
// for _ in 0..k {
// x_expanded_vec.push(0.0);
// }
// for val in x_row {
// x_expanded_vec.push(*val);
// }
// }
if q_val <= gamma {
// Regime 1 Ativo: [x, 0]
x_expanded_vec.extend(x_row);
x_expanded_vec.extend(std::iter::repeat_n(0.0, k));
} else {
// Regime 2 Ativo: [0, x]
x_expanded_vec.extend(std::iter::repeat_n(0.0, k));
x_expanded_vec.extend(x_row);
}
}
let x_expanded = Array2::from_shape_vec((n, 2 * k), x_expanded_vec)
.map_err(|e| GreenersError::ShapeMismatch(e.to_string()))?;
// Rodar Fixed Effects para este Gamma
// Isso já cuida da remoção das médias individuais (mu_i)
let fe_res = FixedEffects::fit(y, &x_expanded, id_slice);
if let Ok(model) = fe_res {
// Calcular SSR do modelo
// SSR = Sigma^2 * df_resid (revertendo a conta do sigma)
let ssr = model.sigma.powi(2) * (model.df_resid as f64);
if ssr < min_ssr {
min_ssr = ssr;
best_gamma = gamma;
best_params = model.params;
best_r2 = model.r_squared;
}
}
}
// 3. Separar os parâmetros
let params_regime1 = best_params.slice(s![0..k]).to_owned();
let params_regime2 = best_params.slice(s![k..2 * k]).to_owned();
Ok(ThresholdResult {
threshold_gamma: best_gamma,
params_regime1,
params_regime2,
r_squared: best_r2,
ssr_min: min_ssr,
n_search: candidates.len() / step,
})
}
}