Monthly DAG
Narrative edges (strongest first)
| From | | To | Lag | β (z) | Child R² |
| public_debt_gdp | → | debt_gap | t | +1.000 | 0.044 |
| nominal_deficit | → | primary_balance | t | +0.925 | 0.821 |
| nominal_deficit | → | public_debt_gdp | t | -0.726 | -9.764 |
| copper | → | tot | t | -0.692 | -17617.685 |
| primary_balance | → | public_debt_gdp | t | -0.688 | -9.764 |
| primary_balance | → | debt_gap | t | -0.688 | 0.044 |
| copper | → | tot_gap | t | -0.661 | -125.773 |
| debt_gap | → | embi_bra | t | +0.553 | -2.291 |
| public_debt_gdp | → | embi_bra | t | +0.553 | -2.291 |
| tot | → | current_account | t | -0.507 | -0.060 |
| tot_gap | → | current_account | t | -0.441 | -0.060 |
| current_account | → | reserves | t | +0.427 | -0.109 |
| dxy | → | embi_bra | t | +0.415 | -2.291 |
| current_account | → | embi_bra | t | +0.332 | -2.291 |
| tot | → | reserves | t | -0.307 | -0.109 |
| copper | → | embi_bra | t | +0.255 | -2.291 |
| tot_gap | → | primary_balance | t | +0.185 | 0.821 |
| reserves | → | reserves_growth | t | +0.177 | 0.048 |
| tot | → | embi_bra | t | -0.147 | -2.291 |
| tot | → | primary_balance | t | +0.121 | 0.821 |
| tot_gap | → | embi_bra | t | +0.111 | -2.291 |
| current_account | → | reserves_growth | t | -0.086 | 0.048 |
| reserves_growth | → | embi_bra | t | +0.056 | -2.291 |
| us_hy_oas | → | embi_bra | t | -0.051 | -2.291 |
| primary_balance | → | embi_bra | t | -0.030 | -2.291 |
ΔEMBI engine (holdout score)
Ridge on ΔEMBI: 1 risk (HY or VIX or S&P) + 1 rates/USD + 1 commodity
+ 1 local + 1 stock. Ranked by holdout R², not quintile lift.
Best: theory|lags=[0] · parents: us_hy_oas, dxy, copper, reserves_growth, public_debt_gdp
· 182 train / 19 test through 2026-09-30.
| From | | To | Lag | β (z) | Child R² |
| us_hy_oas | → | embi_bra | t | +0.482 | 0.586 |
| dxy | → | embi_bra | t | +0.276 | 0.586 |
| reserves_growth | → | embi_bra | t | +0.059 | 0.586 |
| copper | → | embi_bra | t | -0.044 | 0.586 |
| public_debt_gdp | → | embi_bra | t | +0.000 | 0.586 |
| Method | Lags | OOS R² | σ-quintile | Lift vs naive | Naive middle | ±1 bin | 3-class | Sign | RMSE Δ | Edges | Parents |
| theory|lags=[0] ← best | [0] | 0.242 | 52.6% | 0.053 | 47.4% | 89.5% | 68.4% | 57.9% | 0.149 | 9 | 5 |
| pc|lags=[0] | [0] | 0.242 | 52.6% | 0.053 | 47.4% | 89.5% | 68.4% | 57.9% | 0.149 | 16 | 5 |
| pc_strict|lags=[0] | [0] | 0.242 | 52.6% | 0.053 | 47.4% | 89.5% | 68.4% | 57.9% | 0.149 | 14 | 5 |
| tan|lags=[0] | [0] | 0.242 | 52.6% | 0.053 | 47.4% | 89.5% | 68.4% | 57.9% | 0.149 | 9 | 5 |
| hybrid|lags=[0] | [0] | 0.242 | 52.6% | 0.053 | 47.4% | 89.5% | 68.4% | 57.9% | 0.149 | 15 | 5 |
| theory|lags=[0, 1] | [0, 1] | 0.208 | 52.6% | 0.053 | 47.4% | 89.5% | 68.4% | 68.4% | 0.152 | 9 | 10 |
| pc|lags=[0, 1] | [0, 1] | 0.208 | 52.6% | 0.053 | 47.4% | 89.5% | 68.4% | 68.4% | 0.152 | 16 | 10 |
| pc_strict|lags=[0, 1] | [0, 1] | 0.208 | 52.6% | 0.053 | 47.4% | 89.5% | 68.4% | 68.4% | 0.152 | 14 | 10 |
| tan|lags=[0, 1] | [0, 1] | 0.208 | 52.6% | 0.053 | 47.4% | 89.5% | 68.4% | 68.4% | 0.152 | 9 | 10 |
| hybrid|lags=[0, 1] | [0, 1] | 0.208 | 52.6% | 0.053 | 47.4% | 89.5% | 68.4% | 68.4% | 0.152 | 15 | 10 |
| theory|lags=[0, 1, 2] | [0, 1, 2] | 0.146 | 42.1% | -0.053 | 47.4% | 89.5% | 57.9% | 73.7% | 0.158 | 9 | 15 |
| pc|lags=[0, 1, 2] | [0, 1, 2] | 0.146 | 42.1% | -0.053 | 47.4% | 89.5% | 57.9% | 73.7% | 0.158 | 16 | 15 |
| pc_strict|lags=[0, 1, 2] | [0, 1, 2] | 0.146 | 42.1% | -0.053 | 47.4% | 89.5% | 57.9% | 73.7% | 0.158 | 14 | 15 |
| tan|lags=[0, 1, 2] | [0, 1, 2] | 0.146 | 42.1% | -0.053 | 47.4% | 89.5% | 57.9% | 73.7% | 0.158 | 9 | 15 |