TL;DR
This paper introduces a novel case-based decision theory (CBDT) approach to quantify and manage "general uncertainty" in portfolio selection, where neither states of nature nor their probabilities are known. By modeling investor belief in conventional strategies, they identify distinct successful strategies that mimic behavioral biases and adapt to market efficiency, offering a richer understanding of investment decision-making under uncertainty.
What Problem Does This Paper Address?
Traditional economic literature often distinguishes between "risk" (known states, unknown probabilities) and "uncertainty" (known states, unknown probabilities). However, a more challenging scenario is "general uncertainty," where even the states of nature are unknown or difficult to define. This is particularly relevant in portfolio selection, where investors may struggle to envision all possible future market conditions or validate the assumptions of their chosen models. Existing Expected Utility (EU) frameworks are ill-suited for such situations. This paper aims to fill this gap by applying the Case-Based Decision Theory (CBDT), originally developed by Gilboa and Schmeidler, to model and manage this "general uncertainty" in portfolio selection.
Methods
The core of the paper is the development of a CBDT framework to quantify general uncertainty. This is operationalized by defining "general uncertainty" as the weakness of an investor's belief in a conventional portfolio procedure. This belief is represented by a "model belief degree," denoted by $\theta_t$, ranging from 0 (complete disbelief) to 1 (complete belief).
The authors propose a two-stage process:
- Conventional Portfolio Rule: An investor first applies a standard portfolio selection rule (e.g., Markowitz) to determine initial portfolio weights.
- Case-Based Reasoning for Belief Degree: The investor then uses CBDT to dynamically adjust their model belief degree ($\theta_t$). This involves:
- Hypothetical Reasoning: The investor considers all possible values of $\theta_t$ and calculates the hypothetical utility for each.
- Similarity Function: A similarity function (specifically, a modulus similarity function) is used to measure the distance between the current situation and past experiences. This function incorporates an "indifference parameter" ($\kappa$) that dictates how strongly the investor differentiates between acts.
- Aspiration Level: An "aspiration level" ($H_t$) represents the investor's satisfaction threshold, differentiating favorable from unfavorable outcomes. This level is updated endogenously based on past outcomes, reflecting a learning process.
- Objective Function: The investor chooses the $\theta_t$ that maximizes a case-based objective function, which is a weighted sum of hypothetical excess utilities, where weights are determined by the similarity function.
This $\theta_t$ is then used to adjust the conventional portfolio weights, effectively moving the portfolio allocation along the capital market line towards the risk-free asset as belief weakens.
Datasets
The empirical study utilizes weekly returns data for US stocks from the Dow Jones Industrial Average (DJIA) and the US 3-month T-Bill rate as the risk-free rate. Two periods are analyzed:
- Full Sample: January 1, 1973, to October 6, 2005 (1709 observations, 24 stocks).
- Recent Sub-sample: January 1, 1987, to October 6, 2005 (980 observations, 30 stocks).
The data is sourced from DataStream. No new benchmarks are introduced; rather, the study compares CBDT strategies against established non-CBDT approaches.
Key Insights and Ideas
- General Uncertainty as Belief Weakness: The paper reframes general uncertainty not as a lack of information about states, but as a lack of confidence in the validity or applicability of a chosen decision model.
- Two Successful Case-Based Strategies: The study identifies two distinct successful CBDT strategies:
- Fundamental Investor: Characterized by a high indifference parameter ($\kappa > 1$) and near-perfect memory ($\lambda \approx 1$). This investor makes gradual adjustments to their belief degree, relying heavily on historical information and exhibiting biases like anchoring and conservatism. This strategy is more successful in less efficient markets with higher predictability.
- Active Trader: Characterized by a low indifference parameter ($\kappa < 1$) and significant memory ($\lambda \approx 0.5$ to $1$). This investor reacts quickly to new information, exhibiting biases like representativeness and availability. This strategy thrives in more efficient markets.
- Market Efficiency and Strategy Choice: The findings suggest that as markets become more efficient (less predictable), the benefits of fundamental strategies diminish, while active trading strategies become more advantageous. This implies that CBDT can capture evolving market dynamics.
- Behavioral Phenomena: The identified successful strategies naturally incorporate behavioral biases, suggesting that CBDT can provide a framework for understanding how these biases manifest in investment decisions under uncertainty.
Key Results
The empirical results show that:
- Purely Markowitz-based approaches often underperform, even compared to the risk-free asset, especially in the recent period.
- CBDT-adjusted strategies, particularly those based on the Capital Market Line (CML) and constrained Markowitz, can significantly outperform non-CBDT approaches for certain parameter combinations ($\kappa, \lambda$) and investor risk aversion levels ($\gamma$).
- For investors with low risk aversion ($\gamma=2$), the CML approach with good memory and a concave similarity function ($\kappa > 1$) performs well.
- For investors with higher risk aversion ($\gamma=10, 25$), CBDT-adjusted Markowitz rules with a convex similarity function and varying memory levels show superior performance.
- The study provides evidence of decreasing predictability in the US market from 1973-1986 to 1987-2005, supporting the idea of increasing market efficiency. This trend aligns with the observed shift in the success of fundamental versus active trading strategies.
Limitations and Future Work
- Model Specificity: The study focuses on specific portfolio selection rules and a particular similarity function. The generalizability to other rules or similarity functions might require further investigation.
- Behavioral Interpretation: While the paper links strategies to behavioral biases, a deeper exploration of the psychological underpinnings of these case-based strategies could be beneficial.
- Future Work: The authors suggest that the findings shed light on the consequences of applying CBDT to model uncertainty assessment in portfolio selection. Further research could explore the application of these case-based decision patterns to other economic decision-making scenarios or investigate the impact of different types of general uncertainty. The paper also notes that the model could be extended to consider "overconfidence" ($\theta_t > 1$) and "negative belief" ($\theta_t < 0$) in practical situations.
Summary updated