Applications

The project runs seven applications on the Smets and Wouters (2007) model (SW07), all from the 2022Q2 inflation peak or over the full history, each in a rational expectations (RE) version and a learning version. They share the project baseline and differ only in the lines their scenario files set.

Application	Question	Main option	Scenario folder	Runner	Time
1. Peak forecast	How fast does inflation fall from the peak with zero shocks?	expectations.mode	scenarios/sw07/app1_peak/	app1	~30 s
2. Shock replay	Feed in the shocks the RE model says hit the economy: what does the learning economy make of them?	shocks.source	app2_replay/	app2	~10 s
3. Policy deviations	Hold the rate 100 basis points (bp) below or above the Taylor rule for a year	conditions	app3_policy/	app3	~30 s
4. Ramsey policy	Optimal policy under commitment against the rule, four sets of loss weights	policy.*	app4_ramsey/	app4	~2.5 min
5. Density forecasts	How uncertain is the forecast with mark-up or demand shocks?	density.*	app5_density/	app5	~14 min
6. History and decompositions	What history does the learning model back out, and which shocks drove inflation?	history.mode, hetlearn.decompose	app6_history/	app6	~40 s
7. Further expectation rules	How do seven rules from the bounded-rationality literature change Application 1?	expectations.types, expectations.transforms	br_portfolio/	run_br_app1('run')	~2 min

Five model variants change one assumption of Application 1. They are described on the option pages:

Variant	Change	Page
var1_re_variant/v1_mc_learn	model-consistent instead of anchored RE agents	RE agents
var2_plm/v2_*	AR(2), AR(1) without constant, regression and VAR(1) perceived laws of motion	Perceived laws of motion
var3_allvars/v3_*	learning on all twelve forward-looking variables, under each facility	Projection facility
var4_gain/v4_*	constant and decreasing gains against the endogenous gain	Gains
var5_switching/v5_*	intensity of choice 0.5 or 2.25, memory 0.75 or 0.95	Shares and switching

The recursive forecast evaluation (scenarios/sw07/eval/, group eval, about 2.7 hours of compute) is on the Forecast evaluation page.

Running them

setup_paths
run_applications('app3')                 % one group
run_applications({'app1', 'app2'})       % several groups, in this order
run_applications('a5_markup_learn')      % one scenario, logged under its group
run_applications('all')                  % every group, the evaluation included (~3 hours)

Every run writes results/<id>/ and adds a row to results/run_log.csv. A run that fails is printed, logged as error and skipped; the next one runs. help run_applications gives the group table. Application 7 has its own runner, run_br_app1 (Application 7).

Results used on these pages

All numbers on the application pages come from the summary tables of the baseline run (results/summary_v3/). Inflation, output and consumption growth are in % quarter on quarter, annualized, the policy rate in % a year, differences in percentage points (pp).

Checks and reproduction

The summary script results/summary_v3/build_summary_v3.m checks the application runs and writes every check, with its tolerance, to results/summary_v3/checks_v3.txt. For Application 1:

• The RE run a1_re equals a hand-written RE version of SW07 with t-1 information (tests/models/sw07_re_lagged_pinf.model), filtered and simulated on its own, to 1.7e-12. That model was also checked against Dynare.
• In every forecast quarter of a1_learn the share in force equals the share computed at the end of the quarter before, so the projection facility never kept the shares.
• Every run completed, and the facility acted in no Application 1 run.

The checks on the other applications are stated on their pages. To reproduce every result:

run_applications('all')                         % every group; about 3 hours, mostly the evaluation
run('results/summary_v3/build_summary_v3.m')    % summary tables and checks, about 3 minutes

The summary script reads the results of every scenario, so it needs all groups, not only app1.