Dividend Optimizer (Forward-Yield + Entropy)

Overview

This page documents two dividend-focused optimizers available in the backend:

Forward-Yield Allocator (share-based): builds a discrete-share portfolio to maximize expected cash yield subject to risk and allocation caps. It uses raw Close prices (no auto-adjust) and TTM dividends to estimate forward yield per security.
Entropy-regularized Optimizer (continuous weights): maximizes a weighted combination of portfolio dividend yield and portfolio entropy to promote diversification, with an optional volatility cap.
$H(\mathbf{w}) = -\sum_i w_i \log w_i$

Forward yields are computed from TTM (trailing-twelve-month) dividends over a rolling window and the most recent raw Close prices.

• stocks: list of selections, each with ticker and exchange (NSE/BSE) to identify the security.

• budget: optional INR amount; when provided, outputs integer share allocation and cash deployment stats.

• entropy_weight (λ): strength of diversification via entropy in the objective (default 0.05).

• price_lookback_days: days of price history used to estimate covariance for risk constraints.

• yield_lookback_days: window length (days) for aggregating TTM dividends used in forward-yield estimation.

• min_weight_floor (ε): lower bound on each weight to ensure feasibility and avoid zero allocations.

• vol_cap: optional annualized volatility limit applied through

\mathbf{w}^T \Sigma \mathbf{w}

.

• use_median_ttm: if true, smooths TTM by taking the 90-day median of the rolling TTM series for stability.

For each ticker $i$ , compute trailing-twelve-month dividends over a rolling window and last close price $P_i$ to estimate forward yield:

y_i = \dfrac{\mathrm{TTM}_i}{P_i}

Portfolio yield with weights $\mathbf{w}$ is:

y_p = \sum_i w_i y_i

Here $y_i$ denotes the individual forward yield of asset $i$ , and $y_p$ denotes the portfolio forward yield.

With a risk cap (optional):

\mathbf{w}^T \Sigma \mathbf{w} \le \sigma^2_{max}, \quad \sum_i w_i = 1, \quad w_i \ge \epsilon

Objective (maximize expected dividend yield plus diversification):

\max_{\mathbf{w} \ge 0} \; \mathbf{y}^T \mathbf{w} + \lambda H(\mathbf{w})

Subject to:

\sum_i w_i = 1

(weights sum to one)

w_i \ge \epsilon

(long-only with a minimum floor)

\mathbf{w}^T \Sigma \mathbf{w} \le \sigma^2_{max}

(optional annualized volatility cap)

Here $H(\mathbf{w}) = -\sum_i w_i \log w_i$ and $\Sigma$ is the covariance of daily returns scaled by 252.