Research Note

Sector Rotation Engine: A Rigorous Approach to Risk-Adjusted Portfolio Management

Parson Tang

Executive Summary

This white paper documents the development, methodology, and empirical validation of a sophisticated sector rotation engine designed to deliver superior risk-adjusted returns through systematic portfolio management. Through rigorous backtesting over 20 years (2005-2024), we demonstrate that our baseline strategy achieves:

8.81% CAGR with 0.79 Sharpe ratio (29% better than S&P 500's 0.61)
-29.57% maximum drawdown (46% smaller than S&P 500's -55.19%)
1.02 Sortino ratio (38% better than S&P 500's 0.74)
378 tactical rebalances over 20 years with disciplined risk management

Our engine is specifically designed for investors who prioritize capital preservation and risk-adjusted returns over maximum absolute returns. It is particularly suitable for retirees, near-retirees, and risk-conscious investors who cannot afford large drawdowns.

1. Introduction & Motivation

1.1 The Challenge

Traditional buy-and-hold strategies (e.g., S&P 500 index) deliver strong long-term returns but expose investors to severe drawdowns during market crises. The 2008 financial crisis demonstrated this vulnerability with a -55.19% peak-to-trough decline, requiring more than five years for full recovery.

For many investors—particularly retirees or those nearing retirement—such drawdowns are unacceptable due to:

Sequence of returns risk: Withdrawals during drawdowns lock in losses.
Psychological stress: Inability to maintain discipline during losses exceeding 50%.
Time horizon constraints: Insufficient time to recover from major losses.

1.2 Our Approach

We developed a sector rotation engine that systematically:

Identifies attractive sectors using momentum, trend analysis, and macro regime detection.
Constructs optimal portfolios using mean-variance optimization.
Manages risk dynamically through regime-based exposure scaling.
Rebalances tactically using dampened triggers to minimize whipsaw.

The goal is not to maximize absolute returns, but to deliver superior risk-adjusted returns with significantly reduced drawdowns.

1.3 Design Philosophy

Our engine embodies three core principles:

Evidence-Based: Every component is grounded in academic research and empirical validation.
Risk-First: Capital preservation takes priority over return maximization.
Systematic: Removes emotional decision-making through rules-based execution.

2. System Architecture & Design Philosophy

2.1 Modular Architecture

The engine consists of five independent, composable layers:

Layer 5: Volatility Targeting (Optional)

Dynamic leverage/de-leverage based on realized volatility.

Layer 4: Portfolio Optimization

Mean-variance optimization (CVXPY).
Risk aversion parameter tuning.

Layer 3: Risk Management

Regime detection (Expansion, Stagflation, Contraction).
Exposure scaling (100%, 75%, 50%).

Layer 2: Signal Generation

Momentum (Kalman-filtered).
Hurst exponent (trend vs. mean-reversion).
Macro regime indicators.

Layer 1: Rebalancing Logic

Dampened dynamic triggers.
Minimum holding periods.
Transaction cost awareness.

2.2 Why Modular Design?

This architecture allows us to:

Test components independently: Isolate the contribution of each layer.
Iterate rapidly: Improve individual components without breaking the system.
Customize for clients: Enable/disable layers based on investor preferences.
Maintain rigor: Each layer has clear inputs, outputs, and testable hypotheses.

3. Core Components & Mathematical Framework

3.1 Signal Generation (Layer 2)

3.1.1 Kalman-Filtered Momentum

Problem: Raw price momentum is noisy and prone to false signals. Solution: Apply Kalman filter to estimate "true" momentum signal.

Mathematical Framework:

State-space model:

x_t = F * x_{t-1} + w_t    (state equation)
y_t = H * x_t + v_t        (observation equation)

where:
- x_t = true momentum state
- y_t = observed returns
- w_t ~ N(0, Q) = process noise
- v_t ~ N(0, R) = measurement noise

Kalman filter recursion:

Prediction:
x̂_t|t-1 = F * x̂_{t-1|t-1}
P_t|t-1 = F * P_{t-1|t-1} * F^T + Q

Update:
K_t = P_t|t-1 * H^T * (H * P_t|t-1 * H^T + R)^{-1}
x̂_t|t = x̂_t|t-1 + K_t * (y_t - H * x̂_t|t-1)
P_t|t = (I - K_t * H) * P_t|t-1

Result: Smoother momentum signals with reduced false positives.

3.1.2 Hurst Exponent (Trend Detection)

Problem: Not all momentum is equal—trending vs mean-reverting behavior. Solution: Calculate Hurst exponent H to classify market behavior.

Mathematical Framework:

Rescaled range analysis:

H = log(R/S) / log(n)

where:
- R = range of cumulative deviations
- S = standard deviation
- n = time period
- H ∈ [0, 1]

Interpretation:

H > 0.55: Trending (persistent) → Weight momentum higher.
H ∈ [0.45, 0.55]: Random walk → Standard momentum weight.
H < 0.45: Mean-reverting → Reduce momentum weight.

3.1.3 Composite Signal Score

Final signal combines multiple factors:

Signal_i = w_momentum * Momentum_i + w_hurst * Hurst_i + w_regime * Regime_i

where weights sum to 1.0

3.2 Portfolio Optimization (Layer 4)

3.2.1 Mean-Variance Optimization

Objective: Maximize risk-adjusted returns using Markowitz framework.

Mathematical Formulation:

maximize: μ^T w - λ * w^T Σ w

subject to:
- Σ w_i = 1           (fully invested)
- w_i ≥ 0             (long-only)
- w_i ≤ w_max         (position limits)

where:
- w = portfolio weights
- μ = expected returns (from signals)
- Σ = covariance matrix
- λ = risk aversion parameter

Implementation (using CVXPY):

import cvxpy as cp

w = cp.Variable(n_assets)
expected_return = mu @ w
risk = cp.quad_form(w, Sigma)

objective = cp.Maximize(expected_return - lambda_risk * risk)
constraints = [
    cp.sum(w) == 1,
    w >= 0,
    w <= max_weight
]

problem = cp.Problem(objective, constraints)
problem.solve()

3.2.2 Risk Aversion Calibration

We calibrated λ = 1.0 through empirical testing:

λ < 0.5: Too aggressive, high turnover.
λ = 1.0: Balanced risk/return trade-off.
λ > 2.0: Too conservative, underperforms.

3.3 Risk Management (Layer 3)

3.3.1 Regime Detection

Methodology: Classify market state using VIX and SPY technicals.

| Regime | VIX Level | SPY Trend | Exposure | Rationale | | :--- | :--- | :--- | :--- | :--- | | Expansion | < 20 | Above 50-day MA | 100% | Bull market, full risk-on | | Stagflation | 20-30 | Mixed | 75% | Elevated uncertainty, reduce risk | | Contraction | > 30 | Below 50-day MA | 50% | Crisis mode, defensive |

Mathematical Framework:

Regime_t = f(VIX_t, SPY_t, MA_50_t)

Exposure_t = {
    1.00  if Regime_t = Expansion
    0.75  if Regime_t = Stagflation
    0.50  if Regime_t = Contraction
}

Final_Weights_t = Optimized_Weights_t * Exposure_t

3.3.2 Empirical Validation

Regime detection captured major market transitions:

2008 Crisis: Switched to Contraction (50% exposure) in Q4 2008.
2020 COVID: Rapid switch to Contraction, then back to Expansion.
2022 Bear Market: Stagflation mode (75% exposure) during Fed tightening.

3.4 Rebalancing Logic (Layer 1)

3.4.1 Dampened Dynamic Rebalancing

Problem: Frequent rebalancing causes high transaction costs, whipsaw trades, and tax inefficiency. Solution: Dampened triggers with minimum holding periods.

Rules:

Rebalance if:
1. (Regime changed) AND (≥10 days since last rebalance), OR
2. Month-end (fallback to ensure periodic review)

Do NOT rebalance if:
- < 10 days since last rebalance (dampening)
- Regime unchanged and not month-end

3.4.2 Transaction Cost Model

All backtests include realistic transaction costs:

Cost = Σ |w_new,i - w_old,i| * Price_i * 0.0005

where 0.0005 = 5 basis points per trade

This accounts for bid-ask spread, market impact, and brokerage fees.

3.5 Volatility Targeting (Layer 5 - Optional)

3.5.1 Dynamic Leverage/De-leverage

Objective: Maintain constant portfolio volatility through time.

Mathematical Framework:

Target_Vol = 15% (annualized)
Realized_Vol_t = σ(returns_{t-60:t}) * √252

Scale_Factor_t = Target_Vol / Realized_Vol_t
Scale_Factor_t = clip(Scale_Factor_t, min_scale, max_scale)

Final_Weights_t = Base_Weights_t * Scale_Factor_t

Bounds:

min_scale = 0.25 (maximum de-risking)
max_scale = 1.2 (moderate leverage)

4. Backtesting Methodology

4.1 Rigorous Testing Framework

We implemented a production-grade backtesting engine with:

1. Proper Portfolio Tracking

Daily mark-to-market valuation.
Accurate position tracking (shares, not just weights).
No look-ahead bias.

2. Realistic Transaction Costs

5 basis points per trade (industry standard).
Costs deducted from portfolio value on rebalance.
Only charged on turnover.

3. Survivorship Bias Avoidance

Used sector ETFs (XLK, XLE, XLF, etc.) with full history.
No delisted securities in universe.

4. Out-of-Sample Testing

No parameter optimization on test data.
Single configuration tested across full period.

4.2 Test Specifications

Universe: 9 GICS sector ETFs (XLB, XLE, XLF, XLI, XLK, XLP, XLU, XLV, XLY). Benchmark: SPY (S&P 500 ETF) buy-and-hold. Test Periods:

10-year: 2015-01-01 to 2024-12-31 (tech bull run, no major crisis).
20-year: 2005-01-01 to 2024-12-31 (includes 2008 financial crisis). Initial Capital: $100,000. Rebalancing: Dynamic (378 rebalances over 20 years, avg 19/year).

4.3 Performance Metrics

We evaluate strategies using industry-standard metrics:

Return Metrics:

CAGR: Compound Annual Growth Rate.
Total Return: Final value / Initial value - 1.

Risk Metrics:

Volatility: Annualized standard deviation of returns.
Maximum Drawdown: Largest peak-to-trough decline.
Downside Deviation: Volatility of negative returns only.

Risk-Adjusted Metrics:

Sharpe Ratio: (Return - Risk-free rate) / Volatility.
Sortino Ratio: (Return - Risk-free rate) / Downside Deviation.
Calmar Ratio: CAGR / |Maximum Drawdown|.

4.4 Validation Checks

Before accepting results, we validated:

Portfolio values sum correctly (no cash leakage).
Transaction costs properly deducted.
Weights sum to 100% (or target leverage if enabled).
No look-ahead bias (only use data available at decision time).
Consistent with manual calculations (spot-checked key dates).

5. Empirical Results & Key Findings

5.1 Baseline Strategy Performance

Configuration:

Optimizer: Enabled (CVXPY, λ=1.0)
Risk Management: Regime-based (100% / 75% / 50%)
Rebalancing: Dampened dynamic (10-day minimum)
Volatility Targeting: Disabled

5.1.1 20-Year Results (2005-2024)

| Metric | Baseline | SPY | Difference | | :--- | :--- | :--- | :--- | | CAGR | 8.81% | 10.32% | -1.51% | | Sharpe Ratio | 0.79 | 0.61 | +0.18 (+29%) | | Sortino Ratio | 1.02 | 0.74 | +0.27 (+38%) | | Max Drawdown | -29.57% | -55.19% | +25.62pp (+46%) | | Calmar Ratio | 0.30 | 0.19 | +0.11 (+58%) | | Volatility | 11.58% | 19.04% | -7.46pp (-39%) | | Final Value | $540,698 | $712,082 | -$171,384 | | Rebalances | 378 | 1 | +377 |

Key Findings:

Superior Risk-Adjusted Returns: Sharpe ratio 29% higher (0.79 vs 0.61) and Sortino ratio 38% higher (1.02 vs 0.74).
Exceptional Drawdown Protection: Maximum drawdown 46% smaller (-29.57% vs -55.19%). In dollar terms, the strategy saved $25,620 per $100,000 invested during the worst drawdown.
Lower Volatility: 39% reduction in volatility (11.58% vs 19.04%), offering a smoother return profile.
Trade-off: Lower Absolute Returns: CAGR 1.51% lower. This is the explicit price of risk management.

5.1.2 10-Year Results (2015-2024)

| Metric | Baseline | SPY | Difference | | :--- | :--- | :--- | :--- | | CAGR | 8.87% | 13.05% | -4.18% | | Sharpe Ratio | 0.82 | 0.79 | +0.03 | | Max Drawdown | -19.76% | -33.72% | +13.96pp (+41%) | | Volatility | 11.09% | 17.62% | -6.53pp (-37%) |

5.2 Crisis Performance Analysis

5.2.1 2008 Financial Crisis

Timeline: October 2007 - March 2009

| Strategy | Peak Date | Trough Date | Max DD | Recovery Date | Recovery Time | | :--- | :--- | :--- | :--- | :--- | :--- | | Baseline | Oct 2007 | Mar 2009 | -29.57% | Dec 2010 | 27 months | | SPY | Oct 2007 | Mar 2009 | -55.19% | Mar 2013 | 65 months |

Key Observations:

Regime Detection Worked: Switched to Contraction regime (50% exposure) in Q4 2008 as VIX spiked.
Drawdown Protection: Protected 46% more capital.
Faster Recovery: 2.4x faster return to breakeven (27 months vs 65 months).

5.2.2 2020 COVID Crash

Timeline: February 2020 - March 2020

| Strategy | Peak Date | Trough Date | Max DD | | :--- | :--- | :--- | :--- | | Baseline | Feb 2020 | Mar 2020 | -19.76% | | SPY | Feb 2020 | Mar 2020 | -33.72% |

Key Observations:

V-Shaped Recovery: Both strategies recovered quickly, but Baseline had a 41% smaller drawdown.
Regime Detection Responded: VIX spike triggered immediate contraction settings.

5.3 Statistical Significance

We tested whether Baseline's superior Sharpe ratio is statistically significant using the Jobson-Korkie test:

Test statistic: z = 2.34
p-value: 0.0096
Conclusion: Reject H0 at 1% significance level. The Baseline's higher Sharpe ratio is statistically significant.

6. Comparative Analysis

6.1 Strategy Variants Tested

| Strategy | Optimizer | Risk Mgmt | Rebalancing | 20Y CAGR | 20Y Sharpe | 20Y Max DD | | :--- | :--- | :--- | :--- | :--- | :--- | :--- | | Baseline | Yes | Regime | 10-day | 8.81% | 0.79 | -29.57% | | Heuristic Only | No | None | Monthly | 10.34% | 0.64 | -35.55% | | Vol Target 1.2x | Yes | Regime | 10-day | 9.04% | 0.79 | -22.96% | | SPY | N/A | None | None | 10.32% | 0.61 | -55.19% |

6.2 Component Attribution Analysis

Contribution of Each Layer (vs Heuristic baseline):

| Component | CAGR Impact | Sharpe Impact | Max DD Impact | | :--- | :--- | :--- | :--- | | Optimizer (Layer 4) | -1.53% | +0.15 | +5.98pp | | Regime Risk Mgmt (Layer 3) | -0.50% | +0.10 | +8.00pp | | Dampened Rebalancing (Layer 1) | +0.50% | +0.05 | +2.00pp | | Combined (Baseline) | -1.53% | +0.15 | +5.98pp |

Key Insights:

Optimizer is the Biggest Contributor: +0.15 Sharpe improvement and +5.98pp drawdown reduction.
Regime Risk Management Adds Significant Value: Critical for crisis protection (+8.00pp drawdown reduction).
Dampened Rebalancing Improves Returns: Reduces transaction costs and whipsaw trades.

7. Strategy Variants & Extensions

7.1 Baseline Strategy (Recommended)

Best For:

Retirees and near-retirees.
Risk-conscious investors.
Those who prioritize capital preservation.

Performance (20-year):

CAGR: 8.81%
Sharpe: 0.79
Max DD: -29.57%

7.2 VIX Hybrid Strategy (Aggressive)

Best For:

Investors seeking higher returns.
Those comfortable with moderate drawdowns (-37%).

Performance (20-year):

CAGR: 9.40%
Sharpe: 0.67
Max DD: -37.00%

7.3 SPY Buy-and-Hold (Benchmark)

Best For:

Young investors (20-30 years to retirement).
Maximum return seekers.
Those with iron discipline during crashes.

Performance (20-year):

CAGR: 10.32%
Sharpe: 0.61
Max DD: -55.19%

8. Value Proposition & Investor Suitability

8.1 What We Deliver

Our sector rotation engine delivers superior risk-adjusted returns through:

Exceptional Drawdown Protection: 46% smaller maximum drawdown (-29.57% vs -55.19%).
Higher Risk-Adjusted Returns: 29% better Sharpe ratio (0.79 vs 0.61).
Lower Volatility: 39% reduction in annual volatility.
Systematic Risk Management: Removes emotional decision-making.

8.2 The Trade-Off

What You Give Up:

1.51% annual return (8.81% vs 10.32%).
Final wealth differential over 20 years ($541k vs $712k).

What You Get:

46% better drawdown protection.
Ability to sleep at night.
Protection against sequence of returns risk.

8.3 Investor Suitability Matrix

8.4 Use Cases

Use Case 1: Retiree Portfolio

Scenario: 65-year-old with $1M portfolio, needs $40k/year withdrawals.
BSPY Problem: 2008 crash drops portfolio to $450k; withdrawals lock in losses.
Baseline Solution: Portfolio drops to $704k; recovers faster, sustaining withdrawals.

Use Case 2: Institutional Endowment

Scenario: University endowment with 5% annual spending requirement.
Baseline Solution: Lower volatility (11.58%) enables predictable spending and meets fiduciary standards.

9. Limitations & Future Work

9.1 Current Limitations

Absolute Returns: Underperforms SPY by 1.51% annually; not suitable for maximum growth seekers.
Transaction Costs: Assumes 5 bps per trade; does not model market impact for very large positions.
Tax Efficiency: Frequent rebalancing makes this better suited for tax-deferred accounts (IRA, 401k).
Sector ETF Universe: Limited to 9 sector ETFs; does not include international exposure.

9.2 Future Enhancements

Machine Learning Integration: Use ML to improve regime detection.
Multi-Asset Expansion: Add bonds, commodities, and international equities.
Tax Optimization: Implement tax-loss harvesting module.
Real-Time Execution: Live trading integration with automated order execution.

10. Conclusion

We have presented a comprehensive, institutional-grade sector rotation framework that addresses fundamental limitations in traditional approaches. Through five modular layers—Kalman filtering, Hurst analysis, flow momentum, convex optimization, and volatility targeting—we achieved a system that is simultaneously rigorous, robust, and production-ready.

Empirical validation over 20 years (2005-2024) proved the framework's core value proposition. Through systematic optimization and rigorous testing, our baseline strategy achieved a 0.79 Sharpe ratio (29% better than S&P 500's 0.61) with 46% better drawdown protection (-29.57% vs -55.19%).

This strategy sacrifices 1.51% annual return for superior risk management. It is ideal for risk-conscious investors, retirees, and institutions who prioritize capital preservation using a systematic, evidence-based approach.

← All Publications