Ugarte-Ruiz (2026)

Panel local projections without fixed-effects: a cumulative-difference OLS estimator, BBVA Research. The paper reports its applications as figures only. The values below are read off them, to about ±0.2.

Cerra–Saxena (2008): currency crises (Fig. 9)

The data are the Mei–Sheng–Shi replication file empirical_CS_f3.csv (local only; see the Mei–Sheng–Shi page). Its outcomes are pre-built long differences cf{h}GRRT_WB = $y_{t+h} - y_{t-1}$ of log real GDP (× 100). The h = 0 outcome is the growth rate $g_t$, which is the lead of l1GRRT_WB. The specification is the crisis dummy, four of its lags and four lags of growth, pooled with an intercept and without country effects:

f = @formula(cf4GRRT_WB ~ CRISIS + l1CRISIS + l2CRISIS + l3CRISIS + l4CRISIS +
                          l1GRRT_WB + l2GRRT_WB + l3GRRT_WB + l4GRRT_WB)
panellp(f, data; id = :cntyNUM, time = :obs, horizon = 0,
        estimator = CDOLS(validate = false))      # a pre-built outcome column
$h$012345678910
panellp−1.07−2.53−3.34−4.10−4.49−4.65−5.18−4.91−5.11−5.32−5.69
Fig. 9, CMOLS−1.1−2.5−3.4−4.1−4.5−4.65−5.2−4.9−5.1−5.3−5.7

Every horizon is within 0.06 of the figure (a local test item checks 0.1).

Mian–Sufi–Verner (2017): household and firm debt (Fig. 10)

Not reproduced. The paper does not state the CD-OLS outcome, its base period or how the outcome lags enter for this application. With the MSS file empirical_MSV_f2.csv, the closest variant tried (outcome $y_{t+h} - y_t$, both debt shocks and four lags of each, four lags of $\Delta y$) matches Fig. 10 up to $h \approx 3$ and drifts away after. The difference is not only the CD-OLS choices: even Fig. 10's fixed-effects and jackknife lines differ from FE and SPJ on the MSS file (household debt at $h = 7$: −0.47 and −0.65, against −0.387 and −0.558), whereas Fig. 9's match the Cerra–Saxena file. So the paper's Mian–Sufi–Verner data or specification differ from the MSS file.

Furceri et al. (2018): tariffs (Fig. 8)

Not attempted: the data are not part of any replication kit available here.