Fix CRRA=1.0 division by zero errors across consumption models

HARK/ConsumptionSaving/ConsBequestModel.py

@@ -50,7 +50,7 @@
     MargValueFuncCRRA,
     ValueFuncCRRA,
 )
-from HARK.rewards import UtilityFuncCRRA, UtilityFuncStoneGeary
+from HARK.rewards import UtilityFuncCRRA, UtilityFuncStoneGeary, vNvrsSlope
 from HARK.utilities import make_assets_grid
 
 
@@ -416,8 +416,8 @@ def calc_vPPnext(S, a, R):
         vNvrsP_temp = vP_temp * uFunc.derinv(v_temp, order=(0, 1))
         mNrm_temp = np.insert(mNrm_temp, 0, mNrmMinNow)
         vNvrs_temp = np.insert(vNvrs_temp, 0, 0.0)
-        vNvrsP_temp = np.insert(vNvrsP_temp, 0, MPCmaxEff ** (-CRRA / (1.0 - CRRA)))
-        MPCminNvrs = MPCminNow ** (-CRRA / (1.0 - CRRA))
+        vNvrsP_temp = np.insert(vNvrsP_temp, 0, vNvrsSlope(MPCmaxEff, CRRA))
+        MPCminNvrs = vNvrsSlope(MPCminNow, CRRA)
         vNvrsFuncNow = CubicInterp(
             mNrm_temp, vNvrs_temp, vNvrsP_temp, MPCminNvrs * hNrmNow, MPCminNvrs
         )

HARK/ConsumptionSaving/ConsGenIncProcessModel.py

@@ -49,6 +49,7 @@
     CRRAutilityP_invP,
     CRRAutilityPP,
     UtilityFuncCRRA,
+    vNvrsSlope,
 )
 from HARK.utilities import make_assets_grid
 
@@ -486,7 +487,7 @@ def calc_vPP_next(S, a, p):
         )
 
         # Add data at the lower bound of p
-        MPCminNvrs = MPCminNow ** (-CRRA / (1.0 - CRRA))
+        MPCminNvrs = vNvrsSlope(MPCminNow, CRRA)
         m_temp = np.reshape(mLvl_temp[:, 0], (aNrmCount + 1, 1))
         mLvl_temp = np.concatenate((m_temp, mLvl_temp), axis=1)
         vNvrs_temp = np.concatenate((MPCminNvrs * m_temp, vNvrs_temp), axis=1)

HARK/ConsumptionSaving/ConsIndShockModel.py

@@ -56,6 +56,7 @@
     CRRAutilityP_invP,
     CRRAutilityPP,
     UtilityFuncCRRA,
+    vNvrsSlope,
 )
 from HARK.utilities import make_assets_grid
 from scipy.optimize import newton
@@ -373,7 +374,7 @@ def solve_one_period_ConsPF(
     # Calculate (pseudo-inverse) value at each consumption kink point
     vNow = uFunc(cNrmNow) + EndOfPrdv
     vNvrsNow = uFunc.inverse(vNow)
-    vNvrsSlopeMin = MPCminNow ** (-CRRA / (1.0 - CRRA))
+    vNvrsSlopeMin = vNvrsSlope(MPCminNow, CRRA)
 
     # Add an additional point to the list of gridpoints for the extrapolation,
     # using the new value of the lower bound of the MPC.
@@ -786,8 +787,8 @@ def solve_one_period_ConsIndShock(
         vNvrsP_temp = vP_temp * uFunc.derinv(v_temp, order=(0, 1))
         mNrm_temp = np.insert(mNrm_temp, 0, mNrmMinNow)
         vNvrs_temp = np.insert(vNvrs_temp, 0, 0.0)
-        vNvrsP_temp = np.insert(vNvrsP_temp, 0, MPCmaxNow ** (-CRRA / (1.0 - CRRA)))
-        MPCminNvrs = MPCminNow ** (-CRRA / (1.0 - CRRA))
+        vNvrsP_temp = np.insert(vNvrsP_temp, 0, vNvrsSlope(MPCmaxNow, CRRA))
+        MPCminNvrs = vNvrsSlope(MPCminNow, CRRA)
         vNvrsFuncNow = CubicInterp(
             mNrm_temp,
             vNvrs_temp,
@@ -1052,8 +1053,8 @@ def solve_one_period_ConsKinkedR(
         vNvrsP_temp = vP_temp * uFunc.derinv(v_temp, order=(0, 1))
         mNrm_temp = np.insert(mNrm_temp, 0, mNrmMinNow)
         vNvrs_temp = np.insert(vNvrs_temp, 0, 0.0)
-        vNvrsP_temp = np.insert(vNvrsP_temp, 0, MPCmaxNow ** (-CRRA / (1.0 - CRRA)))
-        MPCminNvrs = MPCminNow ** (-CRRA / (1.0 - CRRA))
+        vNvrsP_temp = np.insert(vNvrsP_temp, 0, vNvrsSlope(MPCmaxNow, CRRA))
+        MPCminNvrs = vNvrsSlope(MPCminNow, CRRA)
         vNvrsFuncNow = CubicInterp(
             mNrm_temp,
             vNvrs_temp,

HARK/ConsumptionSaving/ConsIndShockModelFast.py

@@ -48,6 +48,7 @@
     CRRAutility_invP,
     CRRAutilityP,
     CRRAutilityP_inv,
+    vNvrsSlope,
     CRRAutilityP_invP,
     CRRAutilityPP,
     cubic_interp_fast,
@@ -405,7 +406,7 @@ def _solveConsPerfForesightNumba(
     mNrmMinNow = mNrmNow[0]
 
     # See the PerfForesightConsumerType.ipynb documentation notebook for the derivations
-    vFuncNvrsSlope = MPCmin ** (-CRRA / (1.0 - CRRA))
+    vFuncNvrsSlope = vNvrsSlope(MPCmin, CRRA)
 
     return (
         mNrmNow,
@@ -936,8 +937,8 @@ def _add_vFuncNumba(
     vNvrsP = vPnow * utility_invP(vNrmNow, CRRA)
     mNrmGrid = _np_insert(mNrmGrid, 0, mNrmMinNow)
     vNvrs = _np_insert(vNvrs, 0, 0.0)
-    vNvrsP = _np_insert(vNvrsP, 0, MPCmaxEff ** (-CRRA / (1.0 - CRRA)))
-    MPCminNvrs = MPCminNow ** (-CRRA / (1.0 - CRRA))
+    vNvrsP = _np_insert(vNvrsP, 0, vNvrsSlope(MPCmaxEff, CRRA))
+    MPCminNvrs = vNvrsSlope(MPCminNow, CRRA)
 
     return (
         mNrmGrid,

HARK/ConsumptionSaving/ConsMarkovModel.py

@@ -39,6 +39,7 @@
     CRRAutilityP_inv,
     CRRAutilityP_invP,
     CRRAutilityPP,
+    vNvrsSlope,
 )
 from HARK.utilities import make_assets_grid
 
@@ -640,10 +641,8 @@ def calc_vPPnext(S, a, R):
             vNvrsP_now = vP_now * uFunc.derinv(v_now, order=(0, 1))
             mNrm_temp = np.insert(mNrm_for_vFunc, 0, mNrmMin_i)  # add the lower bound
             vNvrs_now = np.insert(vNvrs_now, 0, 0.0)
-            vNvrsP_now = np.insert(
-                vNvrsP_now, 0, MPCmaxEff[i] ** (-CRRA / (1.0 - CRRA))
-            )
-            # MPCminNvrs = MPCminNow[i] ** (-CRRA / (1.0 - CRRA))
+            vNvrsP_now = np.insert(vNvrsP_now, 0, vNvrsSlope(MPCmaxEff[i], CRRA))
+            # MPCminNvrs = vNvrsSlope(MPCminNow[i], CRRA)
             vNvrsFuncNow = LinearInterp(
                 mNrm_temp,
                 vNvrs_now,

HARK/ConsumptionSaving/ConsPrefShockModel.py

@@ -34,7 +34,7 @@
     MargMargValueFuncCRRA,
     ValueFuncCRRA,
 )
-from HARK.rewards import UtilityFuncCRRA
+from HARK.rewards import UtilityFuncCRRA, vNvrsSlope
 
 __all__ = [
     "PrefShockConsumerType",
@@ -376,8 +376,8 @@ def calc_vPPnext(S, a, R):
         vNvrsP_temp = vP_temp * uFunc.derinv(v_temp, order=(0, 1))
         mNrm_temp = np.insert(mNrm_temp, 0, mNrmMinNow)
         vNvrs_temp = np.insert(vNvrs_temp, 0, 0.0)
-        vNvrsP_temp = np.insert(vNvrsP_temp, 0, MPCmaxEff ** (-CRRA / (1.0 - CRRA)))
-        MPCminNvrs = MPCminNow ** (-CRRA / (1.0 - CRRA))
+        vNvrsP_temp = np.insert(vNvrsP_temp, 0, vNvrsSlope(MPCmaxEff, CRRA))
+        MPCminNvrs = vNvrsSlope(MPCminNow, CRRA)
         vNvrsFuncNow = CubicInterp(
             mNrm_temp, vNvrs_temp, vNvrsP_temp, MPCminNvrs * hNrmNow, MPCminNvrs
         )
@@ -683,8 +683,8 @@ def calc_vPPnext(S, a, R):
         vNvrsP_temp = vP_temp * uFunc.derinv(v_temp, order=(0, 1))
         mNrm_temp = np.insert(mNrm_temp, 0, mNrmMinNow)
         vNvrs_temp = np.insert(vNvrs_temp, 0, 0.0)
-        vNvrsP_temp = np.insert(vNvrsP_temp, 0, MPCmaxEff ** (-CRRA / (1.0 - CRRA)))
-        MPCminNvrs = MPCminNow ** (-CRRA / (1.0 - CRRA))
+        vNvrsP_temp = np.insert(vNvrsP_temp, 0, vNvrsSlope(MPCmaxEff, CRRA))
+        MPCminNvrs = vNvrsSlope(MPCminNow, CRRA)
         vNvrsFuncNow = CubicInterp(
             mNrm_temp, vNvrs_temp, vNvrsP_temp, MPCminNvrs * hNrmNow, MPCminNvrs
         )

HARK/ConsumptionSaving/ConsRiskyAssetModel.py

@@ -40,7 +40,7 @@
     MargValueFuncCRRA,
     ValueFuncCRRA,
 )
-from HARK.rewards import UtilityFuncCRRA
+from HARK.rewards import UtilityFuncCRRA, vNvrsSlope
 from HARK.utilities import make_assets_grid
 
 ###############################################################################
@@ -1557,8 +1557,8 @@ def calc_vPPnext(S, a):
         vNvrsP_temp = vP_temp * uFunc.derinv(v_temp, order=(0, 1))
         mNrm_temp = np.insert(mNrm_temp, 0, mNrmMinNow)
         vNvrs_temp = np.insert(vNvrs_temp, 0, 0.0)
-        vNvrsP_temp = np.insert(vNvrsP_temp, 0, MPCmaxEff ** (-CRRA / (1.0 - CRRA)))
-        # MPCminNvrs = MPCminNow ** (-CRRA / (1.0 - CRRA))
+        vNvrsP_temp = np.insert(vNvrsP_temp, 0, vNvrsSlope(MPCmaxEff, CRRA))
+        # MPCminNvrs = vNvrsSlope(MPCminNow, CRRA)
         vNvrsFuncNow = CubicInterp(mNrm_temp, vNvrs_temp, vNvrsP_temp)
         vFuncNow = ValueFuncCRRA(vNvrsFuncNow, CRRA)
     else:

HARK/ConsumptionSaving/ConsWealthPortfolioModel.py

@@ -36,20 +36,13 @@
     make_lognormal_kNrm_init_dstn,
     make_lognormal_pLvl_init_dstn,
 )
-from HARK.rewards import UtilityFuncCRRA
+from HARK.rewards import (
+    UtilityFuncCRRA,
+    CRRAWealthUtilityP,
+)
 from HARK.utilities import NullFunc, make_assets_grid
 
 
-def utility(c, a, CRRA, share=0.0, intercept=0.0):
-    w = a + intercept
-    return (c ** (1 - share) * w**share) ** (1 - CRRA) / (1 - CRRA)
-
-
-def dudc(c, a, CRRA, share=0.0, intercept=0.0):
-    u = utility(c, a, CRRA, share, intercept)
-    return u * (1 - CRRA) * (1 - share) / c
-
-
 def calc_m_nrm_next(shocks, b_nrm, perm_gro_fac):
     """
     Calculate future realizations of market resources mNrm from the income
@@ -299,7 +292,9 @@ def solve_one_period_WealthPortfolio(
     cNrm_now = np.insert(cNrm_now, 0, 0.0)
     cFuncNow = LinearInterp(mNrm_now, cNrm_now)
 
-    dudc_now = dudc(cNrm_now, mNrm_now - cNrm_now, CRRA, WealthShare, WealthShift)
+    dudc_now = CRRAWealthUtilityP(
+        cNrm_now, mNrm_now - cNrm_now, CRRA, WealthShare, WealthShift
+    )
     dudc_nvrs_now = uFunc.derinv(dudc_now, order=(1, 0))
     dudc_nvrs_func_now = LinearInterp(mNrm_now, dudc_nvrs_now)
 

HARK/ConsumptionSaving/ConsWealthUtilityModel.py

@@ -50,7 +50,12 @@
 from HARK.ConsumptionSaving.ConsGenIncProcessModel import (
     GenIncProcessConsumerType,
 )
-from HARK.rewards import UtilityFuncCRRA, CRRAutility
+from HARK.rewards import (
+    UtilityFuncCRRA,
+    CRRAutility,
+    CRRAWealthUtility,
+    CRRAWealthUtilityP,
+)
 from HARK.utilities import NullFunc, make_assets_grid
 
 
@@ -218,16 +223,6 @@ def make_ChiFromOmega_function(CRRA, WealthShare, ChiFromOmega_N, ChiFromOmega_b
     )
 
 
-def utility(c, a, CRRA, share=0.0, intercept=0.0):
-    w = a + intercept
-    return (c ** (1 - share) * w**share) ** (1 - CRRA) / (1 - CRRA)
-
-
-def dudc(c, a, CRRA, share=0.0, intercept=0.0):
-    u = utility(c, a, CRRA, share, intercept)
-    return u * (1 - CRRA) * (1 - share) / c
-
-
 def calc_m_nrm_next(shocks, a_nrm, G, R):
     """
     Calculate future realizations of market resources mNrm from the income
@@ -371,7 +366,7 @@ def solve_one_period_WealthUtility(
     m_temp = aXtraGrid.copy() + mNrmMinNow
     c_temp = cFuncNow(m_temp)
     a_temp = m_temp - c_temp
-    dudc_now = dudc(c_temp, a_temp, CRRA, WealthShare, WealthShift)
+    dudc_now = CRRAWealthUtilityP(c_temp, a_temp, CRRA, WealthShare, WealthShift)
     dudc_nvrs_now = np.insert(uFunc.derinv(dudc_now, order=(1, 0)), 0, 0.0)
     dudc_nvrs_func_now = LinearInterp(np.insert(m_temp, 0, mNrmMinNow), dudc_nvrs_now)
 
@@ -384,7 +379,7 @@ def solve_one_period_WealthUtility(
             calc_v_next, IncShkDstn, args=(a_temp, PermGroFac, Rfree, CRRA, vFuncNext)
         )
         EndOfPrd_v *= DiscFacEff
-        u_now = utility(c_temp, a_temp, CRRA, WealthShare, WealthShift)
+        u_now = CRRAWealthUtility(c_temp, a_temp, CRRA, WealthShare, WealthShift)
         v_now = u_now + EndOfPrd_v
         vNvrs_now = np.insert(uFunc.inverse(v_now), 0, 0.0)
         vNvrsFunc = LinearInterp(np.insert(m_temp, 0, mNrmMinNow), vNvrs_now)

HARK/models/consumer.py

@@ -1,5 +1,6 @@
 from HARK.distributions import Bernoulli, Lognormal, MeanOneLogNormal
 from HARK.model import Control, DBlock, RBlock
+from HARK.rewards import CRRAutility
 
 """
 Blocks for consumption saving problem (not normalized)
@@ -35,7 +36,7 @@
             "p": lambda PermGroFac, p: PermGroFac * p,
             "a": lambda m, c: m - c,
         },
-        "reward": {"u": lambda c, CRRA: c ** (1 - CRRA) / (1 - CRRA)},
+        "reward": {"u": lambda c, CRRA: CRRAutility(c, CRRA)},
     }
 )
 
@@ -52,7 +53,7 @@
             "c": Control(["m"]),
             "a": "m - c",
         },
-        "reward": {"u": lambda c, CRRA: c ** (1 - CRRA) / (1 - CRRA)},
+        "reward": {"u": lambda c, CRRA: CRRAutility(c, CRRA)},
     }
 )
 

HARK/models/consumer.yaml

@@ -30,6 +30,8 @@ blocks:
             a: m - c
 
         reward:
+            # Note: This formula does not support CRRA=1 (log utility).
+            # For CRRA=1, use the Python solvers which handle this case.
             u: c ** (1 - CRRA) / (1 - CRRA)
     - &portfolio_choice
         name: portfolio choice

HARK/models/fisher.py

@@ -3,6 +3,7 @@
 """
 
 from HARK.model import Control, DBlock
+from HARK.rewards import CRRAutility
 
 # This way of distributing parameters across the scope is clunky
 # Can be handled better if parsed from a YAML file, probably
@@ -25,6 +26,6 @@
             "c": Control(["m"]),
             "a": lambda m, c: m - c,
         },
-        "reward": {"u": lambda c: c ** (1 - CRRA) / (1 - CRRA)},
+        "reward": {"u": lambda c: CRRAutility(c, CRRA)},
     }
 )

HARK/models/perfect_foresight.py

@@ -1,5 +1,6 @@
 from HARK.distributions import Bernoulli
 from HARK.model import Control, DBlock
+from HARK.rewards import CRRAutility
 
 # This way of distributing parameters across the scope is clunky
 # Can be handled better if parsed from a YAML file, probably
@@ -28,6 +29,6 @@
             "p": lambda PermGroFac, p: PermGroFac * p,
             "a": lambda m, c: m - c,
         },
-        "reward": {"u": lambda c, CRRA: c ** (1 - CRRA) / (1 - CRRA)},
+        "reward": {"u": lambda c, CRRA: CRRAutility(c, CRRA)},
     }
 )

HARK/models/perfect_foresight_normalized.py

@@ -1,5 +1,6 @@
 from HARK.distributions import Bernoulli
 from HARK.model import Control, DBlock
+from HARK.rewards import CRRAutility
 
 # This way of distributing parameters across the scope is clunky
 # Can be handled better if parsed from a YAML file, probably
@@ -29,6 +30,6 @@
             "c_nrm": Control(["m_nrm"]),
             "a_nrm": lambda m_nrm, c_nrm: m_nrm - c_nrm,
         },
-        "reward": {"u": lambda c: c ** (1 - CRRA) / (1 - CRRA)},
+        "reward": {"u": lambda c: CRRAutility(c, CRRA)},
     }
 )

HARK/numba_tools.py

@@ -9,6 +9,7 @@
     CRRAutilityP_inv,
     CRRAutilityP_invP,
     CRRAutilityPP_X,
+    vNvrsSlope as vNvrsSlope_py,
 )
 
 CRRAutility = njit(CRRAutility_X, cache=True)
@@ -18,6 +19,7 @@
 CRRAutility_invP = njit(CRRAutility_invP, cache=True)
 CRRAutility_inv = njit(CRRAutility_inv, cache=True)
 CRRAutilityP_invP = njit(CRRAutilityP_invP, cache=True)
+vNvrsSlope = njit(vNvrsSlope_py, cache=True)
 
 
 @njit(cache=True, error_model="numpy")