Changes to prepare_buying_goods() are leading to exaggerated growth across all firms

[sternluke]

The recent changes to prepare_buying_goods() are leading to exaggerated growth even with tfp_base_growth_rate = 0.0. I believe the exaggerated growth is coming from two sources:

1. the magnitude is of the investment is large where extra_capital_qty often seems to increase target_capital_inputs by up to ~25% per timestep
2. and the frequency of investment is also high because profitable = hurdle_value > candidate_investment in plan_productivity_investment() is almost always true meaning that target_capital_inputs is increased due to planned_investment on almost all timesteps.

hurdle_rate=0.125 output:

Timestep: 0, Year: 2014, Month:0
unconstrained_target_capital_inputs [10][4] 4.7151e+06
planned_investment [10] 2.4405e+08
weights [10][4] 5.4409e-03
safe_prices [10] 1.1055e+00
extra_capital_qty [10][4] 1.2011e+06
BEFORE: target_capital_inputs [10][4] 4.7151e+06
AFTER APPLYING EXTRA_CAPITAL: target_capital_inputs [10][4] 5.9161e+06

Timestep: 1, Year: 2014, Month:3
unconstrained_target_capital_inputs [10][4] 4.7654e+06
planned_investment [10] 2.3512e+08
weights [10][4] 5.4987e-03
safe_prices [10] 1.0739e+00
extra_capital_qty [10][4] 1.1513e+06
BEFORE: target_capital_inputs [10][4] 4.7654e+06
AFTER APPLYING EXTRA_CAPITAL: target_capital_inputs [10][4] 5.9167e+06

Timestep: 2, Year: 2014, Month:6
unconstrained_target_capital_inputs [10][4] 4.9256e+06
planned_investment [10] 2.2216e+08
weights [10][4] 5.5848e-03
safe_prices [10] 1.0420e+00
extra_capital_qty [10][4] 1.1208e+06
BEFORE: target_capital_inputs [10][4] 4.9256e+06
AFTER APPLYING EXTRA_CAPITAL: target_capital_inputs [10][4] 6.0464e+06

Timestep: 3, Year: 2014, Month:9
unconstrained_target_capital_inputs [10][4] 5.0966e+06
planned_investment [10] 2.1136e+08
weights [10][4] 5.6696e-03
safe_prices [10] 1.0698e+00
extra_capital_qty [10][4] 1.0886e+06
BEFORE: target_capital_inputs [10][4] 5.0966e+06
AFTER APPLYING EXTRA_CAPITAL: target_capital_inputs [10][4] 6.1852e+06

hurdle_rate=0.75 output:

Timestep: 0, Year: 2014, Month:0
unconstrained_target_capital_inputs [10][4] 4.7151e+06
planned_investment [10] 2.4405e+08
weights [10][4] 5.4409e-03
safe_prices [10] 1.1055e+00
extra_capital_qty [10][4] 1.2011e+06
BEFORE: target_capital_inputs [10][4] 4.7151e+06
AFTER APPLYING EXTRA_CAPITAL: target_capital_inputs [10][4] 5.9161e+06

Timestep: 1, Year: 2014, Month:3
unconstrained_target_capital_inputs [10][4] 4.7654e+06
planned_investment [10] 2.3512e+08
weights [10][4] 5.4987e-03
safe_prices [10] 1.0739e+00
extra_capital_qty [10][4] 1.1513e+06
BEFORE: target_capital_inputs [10][4] 4.7654e+06
AFTER APPLYING EXTRA_CAPITAL: target_capital_inputs [10][4] 5.9167e+06

Timestep: 2, Year: 2014, Month:6
unconstrained_target_capital_inputs [10][4] 4.9256e+06
planned_investment [10] 2.2216e+08
weights [10][4] 5.5848e-03
safe_prices [10] 1.0420e+00
extra_capital_qty [10][4] 1.1208e+06
BEFORE: target_capital_inputs [10][4] 4.9256e+06
AFTER APPLYING EXTRA_CAPITAL: target_capital_inputs [10][4] 6.0464e+06

[jose-moran]

Automated Code Analysis (LLM-assisted)

This analysis was generated using automated code review with LLM assistance.

Root Cause Identified

The exaggerated growth in prepare_buying_goods() (firms.py:1285-1326) comes from two issues:

1. Multiple Scaling Factors Compound

Three separate fractions are applied:

• max_cash_fraction (default 0.1): Applied in compute_investment_budget()
• max_investment_fraction (default 0.1): Applied in compute_investment_budget()
• investment_propensity (default 0.2): Applied in plan_productivity_investment()

These may have overlapping intent, leading to compound effects or confusion about which one controls investment magnitude.

2. Annual vs Quarterly Mismatch (Potential)

If the investment fractions (e.g., max_investment_fraction = 0.1 meaning 10% of output) are intended as annual values but applied quarterly, the effective annual rate becomes ~40% instead of 10%.

Impact Calculation

From the issue logs:

extra_capital_qty [10][4] 1.2011e+06
BEFORE: target_capital_inputs [10][4] 4.7151e+06
AFTER: target_capital_inputs [10][4] 5.9161e+06

This is a 25.5% increase per timestep. Over 4 quarters, this compounds to significant growth.

Proposed Fix Options

1. Convert annual to quarterly: Divide investment fractions by 4 for quarterly application
2. Remove duplicate scaling: Choose one of max_cash_fraction, max_investment_fraction, or investment_propensity as the primary control
3. Add explicit time period parameter: Make it clear whether fractions are annual or per-timestep

Luke's Suggested Fix (from discussion)

Setting max_investment_fraction = max_cash_fraction = 0.2/4 (i.e., 5% quarterly = 20% annual) and potentially removing investment_propensity as a modifier produced more reasonable growth rates.

[sternluke]

What do you think about potentially removing investment_propensity as a modifier? Doing so makes sense logically to me but unfortunately it would increase the annual TFP growth beyond the typically observed 1% per year (even removing it is on the big side).

Here is delta TFP with investment_propensity:

Here is delta TFP without it: