vrr/WFGY
2
0
Fork 0
mirror of https://github.com/onestardao/WFGY.git synced 2026-04-26 10:40:55 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 4
WFGY/TensionUniverse/BlackHole/Q101_equity_premium_puzzle.md
PSBigBig 08949e6ca5
Update Q101_equity_premium_puzzle.md
2026-01-31 16:33:55 +08:00

49 KiB
Raw Permalink Blame History

Q101 · Equity premium puzzle

0. Header metadata

ID: Q101
Code: BH_ECON_EQUITY_PREM_L3_101
Domain: Economics
Family: asset_pricing
Rank: S
Projection_dominance: I
Field_type: incentive_field
Tension_type: incentive_tension
Status: Open
Semantics: continuous
E_level: E1
N_level: N1
Last_updated: 2026-01-31

0. Effective layer disclaimer

All statements in this file are made strictly at the effective layer of the Tension Universe (TU) framework.

  • The goal of this document is to specify an effective layer encoding of the equity premium puzzle as it appears in the asset pricing literature.
  • The canonical economic problem is treated as an external input. This file does not claim to prove or disprove that problem, nor to resolve it in the sense used by academic macro-finance.
  • No new theorem about asset pricing, preferences, or macroeconomics is introduced here. All substantive claims about the real economy are assumed to come from the cited literature and from standard empirical summaries.
  • All TU objects in this file (state spaces, observables, mismatch fields, tension scores, counterfactual worlds, invariants) live entirely at the effective layer. They are bookkeeping devices for encoding known models, data summaries, and regularity conditions.
  • This file does not expose any TU deep generative rules, any bottom layer axiomatics, or any explicit mapping from raw micro data into TU internal fields. It only assumes that such mappings could exist for purposes of interpretation.
  • Nothing in this file should be cited as evidence that the equity premium puzzle has been solved, that any particular asset pricing model has been proven correct, or that the real economy must match a given TU encoding.
  • When this document refers to “worlds” or “world-representing states” it only means effective layer models that are compatible with certain observable patterns. It does not make ontological claims about the universe.

1. Canonical problem and status

1.1 Canonical statement

In standard asset pricing, consider a representative investor with time separable preferences, smooth consumption, and a stable attitude toward risk. Let

R_equity_real = long run average real return on a broad equity portfolio
R_safe_real   = long run average real return on a very low default risk asset
EP_obs        = R_equity_real - R_safe_real

Empirical studies for major developed economies, most famously the United States in the twentieth century, report values of EP_obs on the order of several percentage points per year.

In the same models, using plausible values for risk aversion, consumption growth volatility, and other macroeconomic quantities, the model implied equity premium

EP_model = model_implied_equity_premium(gamma, consumption_risk, etc.)

is typically much smaller. For wide ranges of standard parameter values, the gap

EP_gap = EP_obs - EP_model

remains large. Matching the observed EP_obs often requires extreme risk aversion, implausible disaster probabilities, or other features that conflict with other data and with common calibrations.

The equity premium puzzle is the claim that, within the canonical class of models and plausible parameter ranges, the observed equity premium appears too large relative to what those models predict.

This subsection is only a restatement of the canonical formulation as it appears in the literature. It does not modify, extend, or reinterpret the original problem statement.

1.2 Status and difficulty

The equity premium puzzle has remained an open challenge in macro-finance since its formal statement in the mid nineteen eighties. It is not a single theorem with a yes or no answer. It is a persistent mismatch between theory and data that has motivated several research programs.

Partial lines of progress include, among others:

  • Habit formation and non time separable preferences.
  • Rare disaster models and heavy tailed consumption or dividend risks.
  • Incomplete markets and heterogeneous agents with limited risk sharing.
  • Long run risks models, where growth and volatility have persistent components.
  • Market frictions, borrowing constraints, and behavioral departures from full rationality.

None of these approaches has produced a universally accepted resolution that fits the puzzle while also satisfying a broad set of auxiliary constraints. Many approaches can reduce the gap in specific settings, but often at the cost of introducing parameters or mechanisms that are difficult to reconcile with other evidence.

The puzzle remains an organizing problem for research in asset pricing and macro-finance.

1.3 Role in the BlackHole project

Within the BlackHole S-problem collection, Q101 serves as:

  1. The anchor node for macro-finance and asset pricing puzzles where prices, risks, and preferences interact over long horizons.

  2. A prototype for incentive tensions where observed rewards and theoretically required compensation for risk do not align under a given model library and fairness constraints.

  3. A test bed for encoding:

    • long horizon return and consumption data,
    • model implied premia and preference parameters,
    • regularity constraints on economic mechanisms,
    • a structured tension functional that separates “puzzle persists” from “puzzle dissolves” worlds at the effective layer.

References

  1. Rajnish Mehra and Edward C. Prescott, “The Equity Premium: A Puzzle”, Journal of Monetary Economics, 1985.
  2. John H. Cochrane, “Asset Pricing”, Princeton University Press, second edition, 2005.
  3. Rajnish Mehra (editor), “The Handbook of the Equity Risk Premium”, Elsevier, 2008.
  4. National Bureau of Economic Research (NBER), topic pages and working papers on asset pricing and the equity premium, various years.

2. Position in the BlackHole graph

This block records how Q101 sits in the BlackHole graph as a node with edges to other problems. Each edge has a one line reason that refers to components or tension structures defined in this file.

2.1 Upstream problems

These nodes provide foundations, tools, or conceptual frames that Q101 reuses at the effective layer.

  • Q059 (BH_CS_INFO_THERMODYN_L3_059) Reason: supplies information and thermodynamic style constraints reused when interpreting the cost of risk and reward inside the EquityPremium_Tension_Functional.

  • Q091 (BH_EARTH_CLIMATE_SENS_L3_091) Reason: shares a template for representing long horizon uncertainty and tail risks, which informs the construction of RegularityPenalty in Block 3.

  • Q120 (BH_PHIL_VALUE_OF_INFORMATION_L3_120) Reason: provides conceptual tools for how beliefs and information shape incentives, which are imported into the incentive_tension framing of Q101.

2.2 Downstream problems

These nodes directly reuse Q101 components or treat Q101 as a prerequisite.

  • Q104 (BH_ECON_INEQUALITY_DYN_L3_104) Reason: reuses EquityPremium_Tension_Functional and RiskPreferenceConsistency_Observer to model how persistent premia interact with wealth accumulation and inequality.

  • Q105 (BH_COMPLEX_CRASHES_L3_105) Reason: uses MacroFinance_Counterfactual_Template to define crash scenarios where premium dynamics and tail events generate systemic tension.

  • Q121 (BH_AI_ALIGNMENT_L3_121) Reason: adapts the incentive_tension and risk tail structure from Q101 as an analogy for misaligned reward schemes in AI agents.

2.3 Parallel problems

Parallel nodes share similar tension types but do not yet share concrete components.

  • Q102 (BH_ECON_HOME_BIAS_L3_102) Reason: encodes a related incentive_tension between diversification theory and observed portfolio choices, though it focuses on portfolio location rather than premia.

  • Q048 (BH_COSMO_H0_TENSION_L3_048) Reason: mirrors the pattern of a stable gap between model predictions and measured values, expressed as a structured tension, but in cosmology rather than macro-finance.

2.4 Cross domain edges

These edges link Q101 to nodes in other domains that can reuse its components.

  • Q032 (BH_PHYS_QTHERMO_L3_032) Reason: can reuse the way Q101 translates between risk compensation, effective temperature style quantities, and incentive_tension.

  • Q059 (BH_CS_INFO_THERMODYN_L3_059) Reason: shares cross domain methods for connecting economic reward structures to information processing and resource constraints.

  • Q123 (BH_AI_INTERP_L3_123) Reason: uses the decomposition of Tension_EP into mismatch and regularity terms as an interpretability pattern for latent “risk versus reward” circuits in AI models.

3. Tension Universe encoding (effective layer)

All content in this block remains at the effective layer. We describe:

  • state spaces,
  • observable fields and mismatch terms,
  • invariants and tension scores,
  • singular sets and domain restrictions.

We do not describe any hidden TU generative rule or any explicit mapping from raw time series into internal TU fields. Those mappings are treated as external to this file.

3.1 State space

We assume a semantic state space

M_econ

with the following interpretation at the effective layer.

Each state m in M_econ represents a coherent macro-finance configuration consisting of:

  • a choice of asset pricing model from a finite library,
  • a choice of data pack from a finite library of empirical summaries,
  • a vector of long run summary statistics for returns and consumption,
  • a parameter vector capturing the key preference and technology parameters for the chosen model.

Formally, we assume that for each state m there exist

model(m) in L_model
data(m)  in L_data
theta(m) in Theta_model(model(m))

and a summary vector S(m) that contains the relevant long horizon statistics needed by the encoding. We do not describe how S(m) is constructed from raw time series. We only assume that such summaries can be produced in a way that is consistent across the data library.

The encoding class for Q101 consists of:

  • a finite model library L_model,
  • a finite data library L_data,
  • parameter domains Theta_model(M_k) for each model,
  • an explicit functional form for RegularityPenalty,
  • fixed weighting rules w_model, w_data,
  • and a family of selection rules specified at design time.

Any change to these ingredients defines a new encoding class and belongs to a future version of this file.

3.2 Model and data libraries

We define two finite libraries that are fixed at the level of this problem.

  1. Model library
L_model = { M_1, M_2, ..., M_K }

Each M_k is a canonical asset pricing specification, for example:

  • basic CRRA representative agent model,
  • habit formation model,
  • rare disaster augmented model,
  • long run risk model,
  • simple incomplete markets model.

For each M_k we specify a compact parameter domain

Theta_model(M_k)

that includes only parameter values considered economically plausible, such as:

  • risk aversion in a bounded interval [gamma_lo, gamma_hi],
  • consumption growth parameters in ranges consistent with macro data,
  • disaster probabilities and sizes in ranges supported by historical evidence.

The boundaries of these parameter domains are fixed once for Q101 and do not depend on any particular dataset or on the observed equity premium in that dataset. In practice, one can work with a finite grid inside each Theta_model(M_k). That grid, once defined, is also part of the encoding class.

  1. Data library
L_data = { D_1, D_2, ..., D_J }

Each D_j is a data pack that contains:

  • long run summaries of equity and safe asset returns for a given country and period,
  • matching summaries of consumption or income growth,
  • basic metadata about sample length and data quality.

Examples include:

  • United States twentieth century data,
  • post war samples for selected developed economies,
  • global panels of equity premia.

The choice of which countries and periods to include in L_data is fixed at the level of Q101. Enlarging L_data or replacing it by a substantially different set of datasets is considered a change of encoding and would belong to a new version.

  1. Weighting rules and fairness constraints

We introduce nonnegative weights

w_model(k) >= 0,  sum over k w_model(k) = 1
w_data(j)  >= 0,  sum over j w_data(j)  = 1

that specify how models and datasets are aggregated when computing global invariants.

Fairness constraints:

  • L_model, L_data, Theta_model(M_k), w_model, and w_data are chosen without using the size of the equity premium in any specific dataset.
  • No weight, parameter bound, or functional form is allowed to depend on the observed value of EP_obs in an individual dataset.
  • RegularityPenalty, defined below, is treated as part of the encoding class. Its functional form is fixed at design time and cannot be redefined in response to tension outputs.
  • Selection rules that choose representative states for datasets must be specified as general algorithms or procedures that apply uniformly across L_data. They are not allowed to be hand tuned on a per dataset basis.

Within a given version of this file, all these choices are treated as fixed.

3.3 Observables and mismatch fields

We define the following effective observables for each state m in M_econ.

  1. Observed returns
R_equity(m) = long run average equity return summary from data(m)
R_safe(m)   = long run average safe asset return summary from data(m)
EP_obs(m)   = R_equity(m) - R_safe(m)

These are long horizon summaries already contained in S(m).

  1. Model implied equity premium
EP_model(m) = model_implied_premium( model(m), theta(m), S(m) )

This is the equity premium implied by the chosen asset pricing model model(m) with parameters theta(m) and the data summaries in S(m).

  1. Equity premium mismatch
DeltaS_prem(m) = | EP_obs(m) - EP_model(m) |

This is nonnegative and is small when the model matches the observed equity premium for that state.

  1. Preference consistency observable

We define an effective risk aversion observable

gamma_eff(m)

that represents the risk aversion level that the chosen model and state imply in the calibration sense.

We choose fixed constants

gamma_lo > 0
gamma_hi > gamma_lo

representing a plausible range for effective risk aversion, based on external literature and not tuned on Q101 tension outputs.

Preference mismatch is defined as

DeltaS_pref(m) =
  max( 0, gamma_eff(m) - gamma_hi )
+ max( 0, gamma_lo - gamma_eff(m) )

This penalizes parameter choices that imply either extremely low or extremely high effective risk aversion relative to this fixed band.

  1. Regularity penalty

We introduce a nonnegative regularity penalty

RegularityPenalty(m) >= 0

that captures violations of simple macro-finance regularity conditions, such as:

  • implausibly high disaster frequencies or sizes relative to data(m),
  • negative or unstable consumption growth patterns inconsistent with the same data pack,
  • credible violations of basic no arbitrage checks at the summary level,
  • other contradictions with widely accepted constraints in the macro-finance literature.

The precise functional form of RegularityPenalty, including which constraints are checked and how violations are scored, is specified once at the level of Q101 and is part of the encoding class. It depends only on model(m), theta(m), and the summary data in S(m).

3.4 Tension tensor and invariants

At the effective layer, we define the equity premium tension score as

Tension_EP(m) =
  a * DeltaS_prem(m)
+ b * DeltaS_pref(m)
+ c * RegularityPenalty(m)

with constants

a > 0
b > 0
c > 0

fixed once at the level of Q101 and not adjusted per dataset or per model. These constants are chosen according to general criteria described in the TU Encoding and Fairness Charter, such as scale normalization and sensitivity requirements, and are not re tuned to improve the visual appearance of specific numerical results.

We embed this score into a TU style tension tensor

T_ij(m) = S_i(m) * C_j(m) * Tension_EP(m) * lambda(m) * kappa

where:

  • S_i(m) is a source factor representing the strength of the i-th economic signal or assumption in the configuration.
  • C_j(m) is a receptivity factor representing the sensitivity of the j-th downstream component to the equity premium tension.
  • lambda(m) is a convergence state factor imported from the TU core at the effective layer.
  • kappa is a coupling constant for macro-finance incentive tension.

The index sets for i and j and the detailed generation rules for S_i, C_j, and lambda are not specified here. They are treated as abstract knobs whose values are well defined and finite for regular states.

We also define a simple global invariant over the encoding:

I_EP_encoding =
  sum over j [ w_data(j) * Tension_EP( m_star(j) ) ]

where m_star(j) is a representative state for dataset D_j constructed as follows:

  • model(m_star(j)) is selected from L_model,
  • theta(m_star(j)) is chosen in Theta_model(model(m_star(j))),
  • the selection rule is specified at design time, applied uniformly across L_data, and respects the fairness constraints. For example, it may be based on pre committed cross validation procedures, or on a fixed model choice per dataset that does not depend on the size of EP_obs.

This invariant summarizes how much tension remains once we have attempted to reconcile each dataset with the model library under a fixed encoding.

3.5 Singular set and domain restrictions

Some states may fail to produce well defined or finite tension scores. We define the singular set

S_sing = { m in M_econ :
           Tension_EP(m) is undefined
           or not finite
           or any required observable is missing }

We impose the following domain restriction:

  • All equity premium tension analysis is restricted to the regular set

    M_reg = M_econ \ S_sing

  • If an experiment or protocol attempts to evaluate Tension_EP(m) for m in S_sing, the result is treated as out of domain and not as evidence about the underlying economic mechanisms or about the correctness of TU itself.

4. Tension principle for this problem

This block states how Q101 is characterized as a tension problem in the TU framework.

4.1 Core tension principle

At the effective layer, the equity premium puzzle is encoded as a structured gap between:

  1. Observed long run equity premia EP_obs(m).
  2. Model implied premia EP_model(m) from a fixed library of canonical models with plausible parameters.
  3. Regularity conditions on preferences and macro-finance variables, captured by DeltaS_pref and RegularityPenalty.

The core principle is:

  • If there exist states in M_reg that represent the real world and keep Tension_EP(m) small and stable across the model and data libraries under fairness constraints, then the puzzle dissolves at the effective layer for that encoding class.
  • If every such state that reduces DeltaS_prem to a small value forces either large preference mismatch or large regularity penalties, then the puzzle persists as high tension.

The goal is not to force a verdict about which case holds in reality. The goal is to create a precise language that makes the tradeoffs between these outcomes explicit and auditable.

4.2 Low tension worlds

In a low tension world, there exist world representing states m in M_reg such that

DeltaS_prem(m)        is small
DeltaS_pref(m)        is small
RegularityPenalty(m)  is small
Tension_EP(m)         <= epsilon_EP

for a fixed small threshold epsilon_EP that does not grow without bound as we extend the data library or refine parameter grids in ways that respect the encoding class.

Moreover, these low tension states can be selected using rules that are consistent across countries and horizons and that obey the fairness constraints. Different datasets may prefer different models inside L_model, but the pattern of tension values remains coherent once regularity and preference constraints are respected.

4.3 High tension worlds

In a high tension world, for any admissible choice of models, parameters, regularity penalty, and selection rules that respect the fairness constraints, world representing states m in M_reg satisfy

Tension_EP(m) >= delta_EP

for some strictly positive delta_EP that cannot be driven close to zero without either

  • leaving the parameter domains Theta_model(M_k),
  • or inducing very large regularity penalties,
  • or violating the fairness constraints by tuning parameters to individual datasets.

At the level of the encoding, Q101 is the claim that the world belongs either to a low tension sector or to a high tension sector. The effective layer tools in this file are designed to clarify how one would test these alternatives without revealing deep TU generative rules.

5. Counterfactual tension worlds

We now describe two counterfactual worlds purely through patterns of observables and tension scores. They are not ontological claims about the universe. They are templates for how the encoding behaves if certain economic resolutions are available.

5.1 World T (puzzle dissolves)

In World T:

  1. For each major dataset D_j in L_data, there exists at least one state m_j in M_reg with

    Tension_EP(m_j) <= epsilon_EP

    under selection rules that obey the fairness constraints and do not exploit dataset specific parameter tuning.

  2. As we extend the sample period or add comparable countries, the distribution of Tension_EP(m_j) values remains stable and clustered below epsilon_EP, once differences in data quality and sample length are controlled for.

  3. Lower tension is achieved by modest adjustments in model choice and parameters within L_model and Theta_model, without pushing gamma_eff(m) or disaster probabilities into extreme ranges and without triggering large RegularityPenalty.

  4. The global invariant I_EP_encoding stays below a moderate bound:

    I_EP_encoding <= I_EP_low

    even when L_data is enlarged in ways consistent with the encoding.

Interpretation: when models and data are enriched within reasonable and pre committed bounds, the apparent puzzle can be explained away at the effective layer for this encoding class.

5.2 World F (puzzle persists)

In World F:

  1. For many datasets D_j in L_data, every attempt to choose a state m_j in M_reg that makes DeltaS_prem(m_j) small leads to either

    • large DeltaS_pref(m_j), or
    • large RegularityPenalty(m_j).

  2. There exists a positive constant delta_EP such that for a large subset of datasets

    Tension_EP(m_j) >= delta_EP

    for all admissible selections of model(m_j) and theta(m_j) that stay inside the encoding class.

  3. Attempts to reduce tension by changing models or parameters within L_model and Theta_model produce strong instabilities across countries or horizons, rather than a coherent low tension pattern. For example, a parameter choice that lowers tension in one dataset may increase it sharply in others.

  4. The global invariant satisfies

    I_EP_encoding >= I_EP_high

    with I_EP_high > I_EP_low, and this inequality remains robust when reasonable extensions to L_model and L_data are made within the same encoding framework.

Interpretation: the equity premium puzzle reflects a structural mismatch between the canonical model library and real world risk and return patterns. Under this encoding, the mismatch persists as high tension even after systematic efforts to expand the model library inside the given constraints.

5.3 Interpretive note

These counterfactual worlds do not define or construct TU deep fields from raw microeconomic data. They only state that, if such effective layer world models exist, then the observable tension patterns would fall into one of these classes.

Q101 as an effective layer specification does not decide which world we live in. It states the conditions under which one could argue that the puzzle has dissolved or persists, within a clearly bounded encoding class.

6. Falsifiability and discriminating experiments

This block specifies experiments and protocols that can test and potentially falsify the Q101 encoding. They do not solve the economic puzzle. They only evaluate whether this particular encoding behaves in a stable and interpretable way under the TU Encoding and Fairness Charter.

Experiment 1: Cross country stability of equity premium tension

Goal: Test whether the Q101 encoding can maintain a coherent low tension pattern across countries without violating fairness constraints.

Setup:

  • Use a fixed model library L_model and data library L_data as defined in Block 3.
  • Fix Theta_model(M_k), w_model, w_data, and the constants gamma_lo, gamma_hi, a, b, c, epsilon_EP, I_EP_low, and I_EP_high at the level of Q101.
  • These constants and domains are selected according to general principles stated in the TU charters and are treated as part of the encoding class. They are not tuned in response to the outcomes of this experiment.
  • For each dataset D_j, prepare a small set of candidate states m_jk in M_reg corresponding to different models and parameter vectors in Theta_model(M_k).

Protocol:

  1. For each dataset D_j and each candidate state m_jk, compute

    Tension_EP(m_jk)

  2. For each D_j, apply a selection rule that

    • chooses one representative state m_star(j) among the m_jk,
    • uses only information permitted by the fairness constraints, such as cross validation, pre committed model choice for certain dataset groups, or randomization independent of EP_obs,
    • does not adjust w_model or w_data.

  3. Compute the global invariant

    I_EP_encoding =
  sum over j [ w_data(j) * Tension_EP(m_star(j)) ]

  4. Record the distribution of Tension_EP(m_star(j)) across countries.

Metrics:

  • Fraction of datasets D_j with Tension_EP(m_star(j)) <= epsilon_EP.
  • Value of I_EP_encoding.
  • Dispersion of Tension_EP(m_star(j)) across countries and its relation to data quality indicators.

Falsification conditions:

  • If, for any reasonable selection rule that respects the fairness constraints and does not depend on EP_obs in a forbidden way, fewer than a fixed fraction q of datasets can be assigned states with Tension_EP(m_star(j)) <= epsilon_EP, and I_EP_encoding is persistently above I_EP_high, then the current encoding of Tension_EP and the associated parameter bounds is considered falsified at the effective layer.
  • If small bounded changes in Theta_model(M_k) or in the RegularityPenalty functional, still within the pre committed encoding class, produce arbitrarily large and irregular jumps in the pattern of Tension_EP(m_star(j)) across countries, the encoding is considered unstable and rejected.

Semantics implementation note: All observables are treated as real valued quantities on a continuous state space. No discrete or hybrid representation is introduced in this experiment.

Boundary note: Falsifying the Q101 encoding in this way does not solve the canonical equity premium puzzle. It only shows that this specific effective layer encoding is not adequate and should be revised or replaced.

When falsification conditions are met, the correct action is to reject or rewrite the encoding class for Q101. It is not acceptable to silently retune parameter ranges, weights, or penalty functions inside this file solely to repair experimental outcomes.

Experiment 2: Horizon dependence of equity premium tension

Goal: Assess whether the encoding produces reasonable and stable tension patterns when the investment horizon changes.

Setup:

  • Select a subset of datasets in L_data where long run return statistics are available at multiple horizons. For example, five year, ten year, and thirty year averages.
  • For each chosen dataset D_j and each horizon h in {5, 10, 30}, define summary statistics and corresponding states m_j(h) in M_reg.
  • Use the same L_model, Theta_model(M_k), weighting rules, and constants a, b, c, gamma_lo, gamma_hi, epsilon_EP, B_hor as fixed in the encoding class. These choices are not adjusted after inspecting horizon dependent results.

Protocol:

  1. For each dataset D_j and each horizon h, compute

    Tension_EP(m_j(h))

  2. For each D_j, evaluate the range

    Range_j = max over h Tension_EP(m_j(h))
          - min over h Tension_EP(m_j(h))

  3. Summarize the set of Range_j values across all selected datasets.

Metrics:

  • Distribution of Tension_EP(m_j(h)) over horizons.
  • Distribution of Range_j across datasets.
  • Correlation between horizon and average tension level, conditional on data quality.

Falsification conditions:

  • If for a large fraction of datasets the range Range_j exceeds a fixed bound B_hor that was set in advance under the charters, and there is no clear economic explanation for this horizon sensitivity, then the encoding is considered inconsistent with stable long run behavior and is rejected.
  • If reducing Range_j to acceptable levels requires re tuning parameters or regularity penalties beyond the pre set Theta_model domains, or requires horizon specific parameter changes that violate the fairness constraints, the encoding is considered to rely on hidden adjustments and is rejected.

Semantics implementation note: The same continuous field interpretation for economic observables is used at all horizons. The only change is the time horizon over which summary statistics are computed.

Boundary note: Falsifying this aspect of the encoding does not explain the origin of the equity premium puzzle. It only shows that a specific way of encoding horizon dependence is inadequate.

As in Experiment 1, if falsification conditions are triggered, the remedy is to redesign the encoding in a new version, not to retroactively shift parameter bounds or weights inside the current specification.

7. AI and WFGY engineering spec

This block describes how Q101 can be used as an engineering module for AI systems within the WFGY framework, at the effective layer. None of these components assumes that the equity premium puzzle is solved. They are intended to help AI systems reason more coherently about macro-finance tension.

7.1 Training signals

We define training signals that can be used to steer models when they reason about asset pricing and macro-finance.

  1. signal_equity_premium_gap

    • Definition: a nonnegative signal proportional to DeltaS_prem(m) when the model constructs or evaluates macro-finance scenarios.
    • Purpose: penalize internal representations or outputs that imply equity premia far from what the chosen model and parameters would predict, when such models are assumed in the background of a task.

  2. signal_preference_consistency

    • Definition: a signal based on DeltaS_pref(m) that increases when implied risk aversion moves outside [gamma_lo, gamma_hi].
    • Purpose: discourage explanations that fit the equity premium only by invoking extreme risk aversion that conflicts with the encoding.

  3. signal_regular_behavior

    • Definition: a signal derived from RegularityPenalty(m) that increases when disaster probabilities, consumption paths, or other macro quantities become inconsistent with the data library or with basic regularity constraints.
    • Purpose: enforce macro-finance regularity even when the model is trying to match observed premia.

  4. signal_global_tension

    • Definition: a scalar signal equal or proportional to Tension_EP(m) for world representing states.
    • Purpose: provide a compact tension indicator that can be minimized or monitored in tasks where consistency with a given asset pricing framework is part of the background.

7.2 Architectural patterns

We outline module patterns that reuse Q101 structures at the effective layer.

  1. EquityPremium_TensionHead

    • Role: a head that reads internal representations of macro-finance contexts and outputs an estimate of Tension_EP(m).

    • Interface:

      • Inputs: embeddings corresponding to model choice, parameter summaries, and data summaries.
      • Outputs: a scalar tension estimate and its decomposition into DeltaS_prem, DeltaS_pref, and RegularityPenalty.

    • Use: can be trained as an auxiliary task and then used at inference time to detect when an answer relies on an inconsistent equity premium story.

  2. RiskPreferenceConsistency_Observer

    • Role: an observer that infers gamma_eff(m) and related preference summaries from the model’s internal state.

    • Interface:

      • Inputs: internal representations of consumption paths, risk comparisons, and narrative explanations.
      • Outputs: an estimated gamma_eff and a preference mismatch score aligned with DeltaS_pref.

    • Use: can act as a filter to flag answers that implicitly require extreme risk attitudes or inconsistent usage of risk aversion.

  3. MacroFinance_Counterfactual_Gateway

    • Role: a module that makes explicit whether the model is reasoning under a “puzzle dissolves” assumption or a “puzzle persists” assumption.

    • Interface:

      • Inputs: a flag or prompt indicating World T or World F and the relevant macro context.
      • Outputs: constrained internal states that are consistent with the chosen world’s tension pattern.

    • Use: prevents the model from mixing assumptions about the state of the puzzle within a single reasoning chain and helps structure scenario analysis.

7.3 Evaluation harness

We sketch an evaluation harness for AI models that use Q101 components.

  1. Task selection

    • Construct a benchmark of questions involving

      • explanations of the equity premium puzzle itself,
      • tradeoffs between risk and return in different environments,
      • implications of different asset pricing models for long run premia,
      • consequences of extreme risk aversion or disaster scenarios.

  2. Conditions

    • Baseline condition

      • The model answers questions without explicit tension heads or observers. It may still use generic reasoning tools.

    • TU condition

      • The model uses EquityPremium_TensionHead and RiskPreferenceConsistency_Observer to generate auxiliary signals and is trained or constrained to avoid high tension states when the background assumption is that the puzzle is close to being resolved within the canonical model class.

  3. Metrics

    • Consistency of explanations across countries and horizons.
    • Frequency of implicit extreme risk aversion in the explanations.
    • Stability of reasoning when toggling between “puzzle persists” and “puzzle dissolves” prompts.
    • Correct recognition of situations where any strong claim would require a genuine resolution of the equity premium puzzle, in which case the model should mark its answer as speculative.

7.4 Sixty second reproduction protocol

This protocol lets external users experience the effect of the Q101 encoding in a short interaction with an AI system.

  • Baseline setup

    • Prompt: ask an AI system to explain why the equity premium is considered puzzling, including model predictions and observed returns, without mentioning tension, TU, or WFGY.
    • Observation: record whether the explanation mixes incompatible parameter choices, invokes extreme risk aversion without acknowledging the problem, or leaves unexplained gaps between theory and data.

  • TU encoded setup

    • Prompt: ask the same question, but explicitly instruct the system to

      • track EP_obs, EP_model, and Tension_EP,
      • avoid explanations that rely on extreme risk aversion or implausible disaster processes,
      • state clearly which parts of the gap are addressed and which remain unexplained.

    • Observation: record whether the explanation becomes more structured, with clear statements about what is observed, what is modeled, which parameters are being used, and where tension remains.

  • Comparison metric

    • Use a simple rubric that scores

      • clarity about observed versus modeled quantities,
      • explicit treatment of parameter plausibility and regularity,
      • coherence of reasoning when switching between different countries or periods.

  • What to log

    • Prompts and full responses from both setups.
    • Any auxiliary Tension_EP and DeltaS_pref estimates produced by the heads.
    • These logs enable later inspection of the encoding’s behavior without exposing deep TU generative rules.

8. Cross problem transfer template

This block lists reusable components produced by Q101 and the problems that can reuse them at the effective layer.

8.1 Reusable components produced by this problem

  1. ComponentName: EquityPremium_Tension_Functional

    • Type: functional

    • Minimal interface:

      • Inputs: macro_return_summary, model_spec, param_vector
      • Output: tension_value (nonnegative scalar)

    • Preconditions:

      • macro_return_summary comes from a dataset in L_data or from a synthetic world consistent with the same format.
      • model_spec is an element of L_model.
      • param_vector lies in the corresponding Theta_model(model_spec).
      • The functional implements the definition of Tension_EP from Block 3.

  2. ComponentName: RiskPreferenceConsistency_Observer

    • Type: observable

    • Minimal interface:

      • Inputs: consumption_summary, param_vector
      • Outputs: gamma_eff_est, preference_mismatch_score

    • Preconditions:

      • consumption_summary is compatible with the same dataset as macro_return_summary.
      • param_vector lies in the model’s parameter domain.
      • The mismatch score aligns with DeltaS_pref from Block 3.

  3. ComponentName: MacroFinance_Counterfactual_Template

    • Type: experiment_pattern

    • Minimal interface:

      • Inputs: model_class, data_pack
      • Output: a pair {World_T_protocol, World_F_protocol} describing how to test whether tension is reducible or persistent in that context.

    • Preconditions:

      • model_class can produce long run premia and preference summaries.
      • data_pack contains returns and macro quantities needed for Q101 style tension evaluation.

8.2 Direct reuse targets

  1. Q104 (BH_ECON_INEQUALITY_DYN_L3_104)

    • Reused components: EquityPremium_Tension_Functional, RiskPreferenceConsistency_Observer.
    • Why it transfers: wealth and income inequality dynamics depend on the gap between returns to risky and safe assets. The same tension decomposition helps identify regimes where high premia persist and how they feed into inequality.
    • What changes: additional observables for savings behavior, income shocks, and wealth distribution are added to the state space. The core tension functional remains intact and becomes one driver of inequality dynamics.

  2. Q105 (BH_COMPLEX_CRASHES_L3_105)

    • Reused component: MacroFinance_Counterfactual_Template.
    • Why it transfers: systemic crashes can be studied as transitions between low and high tension macro-finance worlds, with Q101 serving as the baseline for normal conditions.
    • What changes: new components for network structure, leverage, and contagion are added. The experiments focus on how small shocks in premia and risk perceptions interact with fragile structures to move the system between low and high tension regimes.

  3. Q121 (BH_AI_ALIGNMENT_L3_121)

    • Reused components: EquityPremium_Tension_Functional (as an analogy) and RiskPreferenceConsistency_Observer.
    • Why it transfers: both asset pricing and AI alignment involve incentive structures where high rewards can be misaligned with underlying risks or constraints.
    • What changes: the economic observables are replaced by reward and safety observables in AI systems. The pattern of decomposing tension into mismatch and regularity terms is preserved as a template for designing and auditing alignment metrics.

9. TU roadmap and verification levels

This block states where Q101 sits in the TU verification ladder and what the next measurable steps are.

9.1 Current levels

  • E_level: E1

    • An effective layer encoding of the equity premium puzzle has been specified, including

      • a finite model library and parameter domains,
      • a finite data library,
      • a precise tension functional,
      • explicit fairness and encoding class constraints,
      • at least two falsifiable experiments.

  • N_level: N1

    • A coherent narrative has been given that explains the puzzle as an incentive tension between observed returns, model predictions, and regularity constraints.
    • Counterfactual worlds and experiments have been outlined in a way that separates economic interpretation from TU internal machinery.

9.2 Next measurable step toward E2

To move from E1 to E2, at least one of the following should be implemented and publicly documented:

  1. A working prototype that

    • instantiates specific examples of L_model, L_data, and Theta_model(M_k),
    • computes Tension_EP(m) for selected states m in M_reg,
    • reports I_EP_encoding and horizon dependence patterns for real data,
    • releases enough detail for external audit of the numeric procedures.

  2. A systematic study of synthetic macro-finance model worlds where

    • some worlds are engineered to have resolving mechanisms for the puzzle,
    • others are constructed to keep high tension by design,
    • the Q101 encoding is tested for its ability to separate these cases without access to the generator labels.

Both steps can be carried out entirely at the effective layer and do not require exposing any deep TU generative rules.

9.3 Long term role in the TU program

In the longer term, Q101 is expected to function as:

  • the reference node for economic and financial incentive tensions involving risk premia and long horizon uncertainty,
  • a calibration site for how TU encodings handle the interaction between data, model libraries, and regularity constraints in complex social systems,
  • a bridge from macro-finance to other domains, such as inequality dynamics, systemic risk, and AI alignment, through reusable tension structures.

10. Elementary but precise explanation

At an elementary level, the equity premium puzzle is about a simple question.

Why do broad stock markets seem to pay so much more than very safe assets, when standard models say people should not need that much extra reward to hold them?

Historically, in many countries, broad stock markets have earned much higher average returns than safe government bonds, even after adjusting for inflation. At the same time, measures of how risky consumption and income are suggest that investors should not require such a large extra return if they behave according to standard textbook models.

In those models, you can adjust a few knobs:

  • how much people dislike risk,
  • how smooth or volatile their consumption is,
  • how often big disasters happen and how severe they are.

Even when you turn these knobs across wide plausible ranges, the model’s predicted equity premium often stays far below what the data show. If you force the model to match the observed premium, you may end up with people who are unrealistically afraid of risk, or with disasters that are far more frequent or severe than history supports.

In the Tension Universe view, instead of trying to prove or disprove a specific economic theorem, we do the following at the effective layer.

  1. We collect a small library of standard asset pricing models and a small library of datasets about returns and consumption.

  2. For each combination of model and data, we compute three things:

    • the observed equity premium,
    • the premium that the model predicts,
    • how extreme the implied preferences and macro assumptions are.

  3. We combine these into a single number called Tension_EP.

This tension number is small when

  • the model’s premium is close to the observed premium,
  • the implied preferences sit in a reasonable range,
  • the macro assumptions stay within regular patterns.

The number is large when you only get a good fit by stretching preferences or macro assumptions beyond what seems credible.

Then we imagine two kinds of worlds.

  • In a “puzzle dissolves” world, as we improve models and data inside the encoding class, we can find configurations where the tension number is small across countries and horizons.
  • In a “puzzle persists” world, every attempt to get a small tension number somewhere makes it large somewhere else, or forces us into very implausible parameter regions or regularity violations.

Q101 does not tell us which world we live in. It does not solve the equity premium puzzle. What it does is

  • provide a precise way to talk about how big the mismatch is,
  • make clear what counts as a fair way to compare models and data in this encoding,
  • create reusable tools that help analyze similar tensions in inequality, systemic risk, and AI alignment.

In this sense, Q101 is the macro-finance prototype for incentive tension problems in the Tension Universe framework.

Tension Universe effective layer footer

This page is part of the WFGY / Tension Universe S-problem collection.

Scope of claims

  • The goal of this document is to specify an effective layer encoding of the named problem, in terms of state spaces, observables, mismatch fields, and tension functionals.
  • It does not claim to prove or disprove the canonical statement restated in Section 1.
  • It does not introduce any new theorem beyond what is already established in the cited literature.
  • It should not be cited as evidence that the corresponding open problem in economics has been solved, or that any specific asset pricing model is correct in the real world.
  • All economic judgments about the size and nature of the equity premium puzzle are assumed to originate in external sources. This file only packages them into the TU language.

Effective layer boundary

  • All objects used here (state spaces M_econ, observables, invariants, tension scores, counterfactual worlds) live at the effective layer of the Tension Universe framework.
  • No TU deep axioms, no bottom layer field equations, and no explicit generative rules from raw micro data to TU internal fields are specified or used here.
  • References to TU core quantities such as lambda(m) and T_ij(m) are purely symbolic. Their detailed construction is outside the scope of this file.
  • Counterfactual “World T” and “World F” are descriptions of possible tension patterns in effective layer models. They are not claims about which world is physically or economically real.

Encoding and fairness boundary

  • The encoding class for Q101 consists of a finite model library L_model, a finite data library L_data, parameter domains Theta_model(M_k), a specified RegularityPenalty functional, fixed weights w_model, w_data, and selection rules that apply uniformly across datasets.
  • These ingredients are chosen at design time under principles stated in the TU charters. They are not adjusted in response to the outcomes of experiments described in this file.
  • Any substantial change in model library, data library, parameter domains, penalty functions, weights, or selection rules corresponds to a different encoding class and should be recorded as a new version of this document.
  • Within a given version, it is not acceptable to retune these elements solely to reduce observed tension or to hide failures of the encoding.

Falsifiability and experiments

  • The experiments in Section 6 are designed to test the internal coherence and stability of the Q101 encoding under the TU Encoding and Fairness Charter.
  • Falsifying this encoding means that the current choice of model library, data library, parameter bounds, penalty functions, and tension functional is not adequate at the effective layer.
  • Falsification of the encoding does not by itself resolve the equity premium puzzle and does not invalidate the canonical economic literature. It only indicates that this particular TU packaging is not satisfactory.
  • Passing these experiments means that the encoding behaves in a stable and interpretable way under the stated constraints. It does not mean that the puzzle is solved.

This page should be read together with the following charters:

Index:
← Back to Event Horizon
← Back to WFGY Home

Consistency note:
This entry has passed the internal formal-consistency and symbol-audit checks under the current WFGY 3.0 specification.
The structural layer is already self-consistent; any remaining issues are limited to notation or presentation refinement.
If you find a place where clarity can improve, feel free to open a PR or ping the community.
WFGY evolves through disciplined iteration, not ad-hoc patching.