Speed

tidyMacro against the reference MATLAB codes, same specifications, same thread budget

Every specification on the Validation page was timed in tidyMacro and in the MATLAB code it is compared with. The data, lag orders, horizons and draw counts are the same on both sides.

Thread budget. Both sides get 3 threads. tidyMacro receives n_threads = 3; MATLAB runs under maxNumCompThreads(3). The reference MATLAB codes have no parfor loops, so in MATLAB the budget only reaches multithreaded linear algebra.
Timing. Wall-clock time of one complete call, including VAR estimation. Each routine is run once untimed as a warm-up. Monte Carlo routines are then timed 3 times with different seeds and deterministic routines 20 times. The reported figure is the median.
Sign + external IV. The full Cesa-Bianchi and Sokol (2022) specification takes hours in the VAR Toolbox, so a lighter version is timed: one instrument (the monetary-policy surprise) instead of two, and the financial-shock signs imposed on impact only.
LP-DiD is not timed, because its reference is Stata output.
Machine. Intel Core i5-7360U (2 cores, 4 threads), macOS 13.6, MATLAB R2025b, R 4.6.0, tidyMacro 1.1.2.

Figure 1: Median wall-clock time per call, log scale. The label on the right of each row is the speed-up, MATLAB time divided by tidyMacro time.

Method	Data	Workload	tidyMacro function	MATLAB (s)	tidyMacro (s)	Speed-up
Median seconds per call
Same data and specification on both sides; 3 threads each
Monte Carlo
Sign + external IV (light)	CBS2022	1,000 accepted draws, one IV, impact signs	fSignRestr(instrument =)	17.8	0.288	62×
Sign + narrative	ADRR2018	1,000 accepted draws, Volcker 1979m10	fSignRestr(narrative =)	54.7	1.02	53×
Long run (Blanchard-Quah)	Gali1999	1,000 residual-bootstrap draws, h = 0-40	fVAR + fBootstrapBQ	7.97	0.152	52×
Sign restrictions	Uhlig2005	1,000 accepted draws, signs over h = 0-5	fSignRestr	27.1	0.610	44×
Short run (Cholesky)	Bloom2009	1,000 wild-bootstrap draws, h = 0-48	fVAR + fBootstrapChol	93.2	2.42	39×
External instrument (MBB)	Kaenzig2021	2,000 moving-block draws, h = 0-48	fVAR + fBootstrapIVMBB	11.7	2.43	4.8×
Deterministic
LP-IV	JordaTaylor2025IV	h = 0-48	fLPIV	0.0344	0.0125	2.7×
LP, exogenous shock	JordaTaylor2025	h = 0-17	fLP	0.00974	0.00500	1.9×
Panel LP	MianSufiVerner2017	h = 0-6, two shocks	fLPPanel	0.0202	0.0125	1.6×
External instrument (MSW)	Kaenzig2021	Weak-IV robust bands, h = 0-48	fVAR + fMSW	0.450	0.545	0.83×