When Rigorous Optimization Fails: The Critical Role of Objective Functions in Systematic Hedging

When Rigorous Optimization Fails: The Critical Role of Objective Functions in Systematic Hedging Ioan Bˆaldea Theoretical Chemistry, Heidelberg University Im Neuenheimer Feld 229, D-69120 Heidelberg, Germany E-mail: [email protected] November 1, 2025 Abstract This pedagogical study presents a comprehensive framework for systematic optimization of hedging trading strategies across diverse market regimes. We demonstrate that traditional parameter selection approaches often yield suboptimal results due to constrained search spaces, while systematic exploration reveals non-intuitive optimal configurations. Using a modified geometric Brownian motion process with regime-specific parameters, we generate synthetic market data across six distinct regimes and test a simultaneous long-short hedging strategy with ATR-based position sizing. Our multi-seed validation approach ensures statistical robustness, revealing that optimal parameters (stop-loss multiplier: 1.37, take-profit multiplier: 1.50) achieve 97.2% hedging success rate, significantly outperforming intuitively selected parameters. This research emphasizes the importance of broad parameter exploration, proper statistical validation, and the fundamental tradeoff between success frequency and profit magnitude in systematic trading. At the same time and even more importantly pragmatically, our analysis reveals a more fundamental methodological insight: successful optimization requires alignment between objective functions and practical goals. While we achieved “attractive” success rates, this study demonstrates how even rigorous optimization can yield practically suboptimal results when objectives mismatch real-world priorities. Because what matters is not frequency of success alone, but the fundamental relationship between profit magnitude and loss magnitude across the strategy’s entire return distribution. Disclaimer: This research represents academic simulation work for educational purposes only. All trading involves substantial risk of loss, and past performance does not guarantee future results. 1 Introduction The development of systematic trading strategies often relies on intuitive parameter selection or limited optimization within artificially constrained ranges. This approach risks mistaking local optima for global solutions and may miss fundamentally different strategy behaviors that emerge only through comprehensive parameter exploration. The problem is particularly acute for hedging strategies, where simultaneous long and short positions create complex interactions that defy intuitive parameterization. Our research addresses these challenges through a reproducible framework that combines: •Systematic parameter optimization across broad search spaces •Multi-market regime testing for strategy robustness •Multi-seed statistical validation for result reliability •Transparent methodology with complete reproducibility 1 We demonstrate that what appears to be an optimal parameter set within conventional ranges may be merely an artifact of constrained exploration, while the true optimum lies in unexpected regions of the parameter space. However, this study also serves as a cautionary tale: even with methodologically sound optimization techniques, selecting an inappropriate optimization objective can lead to technically correct but practically useless results. Our high success-rate optimum demonstrates that rigorous mathematics applied to the wrong goal still produces suboptimal outcomes in real-world terms. 2 Methodology 2.1 Philosophical Framework The core philosophy of this research is that systematic strategy development requires: 1. Broad Exploration: Testing parameter ranges beyond intuitive boundaries 2. Statistical Rigor: Multiple random seeds for reliable inference 3. Regime Awareness: Performance evaluation across market conditions 4. Transparent Reproducibility: Complete methodological disclosure This approach contrasts with common practices that optimize within narrow, intuitively ”reasonable” ranges and draw conclusions from single random realizations. 2.2 Market Data Generation We generate synthetic OHLC market data using a modified geometric Brownian motion process with regimespecific parameters [1, 2]: Pt=Pt−1·exp µregime ·∆t+σregime ·√∆t·t(1) where t∼ N(0,1). Our regime-switching approach follows the financial econometrics literature [3, 4] where parameters vary by market regime: Table 1: Market regime parameters for synthetic data generation Market Regime Trend (µ) Volatility (σ) Mean Reversion (θ) Characteristics Strong Trending 0.0003 0.02 0.0 Strong directional momentum Healthy Trending 0.0002 0.015 0.0 Sustainable trends Weak Trending 0.00005 0.01 0.0 Low momentum environments Ranging 0.0 0.01 0.1 Mean-reverting behavior High Volatility 0.0001 0.03 0.0 Elevated price fluctuations Low Volatility 0.0001 0.003 0.0 Reduced price movements For ranging markets, we incorporate mean reversion through an Ornstein-Uhlenbeck process:[5, 6]: dPt=θ·(µ−Pt)·dt +σ·dWt(2) where θ= 0.1 controls the speed of mean reversion, ensuring price returns to the mean level µ= 100.0 (initial price). This formulation is widely used in financial modeling of mean-reverting assets [7]. 2 2.3 Trading Cost Structure Realistic transaction costs are incorporated throughout all simulations: •Spread: 0.02% of current price, applied on both entry and exit •Slippage: 0.01% of current price, modeling execution imperfections •Swap Costs: Long positions: -0.0001% per period, Short positions: +0.00005% per period •Position Sizing: Fixed 1% of account equity per position ($100 initial risk) •Maximum Open Positions: 2 simultaneous positions total (enforced constraint) These costs ensure strategy performance reflects realistic trading conditions rather than idealized simulations. 2.4 Strategy Implementation Specifications Our core strategy implements the following detailed logic: Entry Conditions: •Signal: RSI-based entry filter (RSI<30 for long, RSI>70 for short) •ATR Calculation: 14-period Average True Range for volatility adjustment •Position Limits: Maximum 2 open positions simultaneously (long + short hedge) •Entry Timing: Next bar execution after signal generation Exit Conditions: •Stop-Loss: Price ±s×ATR14, where sis the stop-loss multiplier •Take-Profit: Price ±p×ATR14, where pis the take-profit multiplier •Constraint:p>sensures positive risk-reward ratio •Execution: Intra-bar checks for stop and take-profit levels Risk Management: •Maximum Drawdown Control: Position sizing limits single-trade risk to 1% of equity •Hedging Logic: Simultaneous long/short positions act as natural risk control •Trade Frequency: Limited by RSI filter and maximum position constraints 2.5 Experimental Design Specifications The comprehensive experimental design includes: •Data Length: 1000 OHLC bars per simulation, representing approximately 2 months of 1 h charts •Initial Capital:$10,000 starting equity •Position Sizing: Fixed fractional (1% of current equity) •Total Simulations: 6 markets ×4 seeds ×400 parameters = 9,600 individual backtests •Average Trade Count:∼50-150 hedging pairs per simulation depending on parameters •Computational Resources: Total simulation time approximately 4-6 hours on standard hardware 3 2.6 Performance Metrics We evaluate strategy performance using multiple complementary metrics: •Primary Metric: Hedge success rate = count(Pnet >0) total hedges •Secondary Metrics: Total PnL, Sharpe ratio, maximum drawdown, profit factor •Risk-Adjusted Measures: Return per unit of risk, consistency metrics •Statistical Significance: Confidence intervals across multiple seeds This comprehensive specification ensures complete reproducibility and provides readers with all necessary details to understand the experimental setup and potential limitations. 2.7 The Importance of Multiple Seeds Financial time series generation involves inherent randomness. Drawing conclusions from single random realizations risks overfitting to specific noise patterns. Our multi-seed approach addresses the inherent randomness in financial time series generation, preventing overfitting to specific noise patterns [11]. This methodology provides statistical significance and ensures robustness across different market realizations: •Quantifies Uncertainty: Measures performance variation across different random generations •Ensures Robustness: Validates that optimal parameters work across different market realizations •Provides Statistical Significance: Enables proper inference about parameter performance •Prevents Overfitting: Avoids optimizing to specific random patterns We use seeds 42, 101, 2025, and 1234 to ensure our findings are not artifacts of particular random number sequences. 2.8 Hedging Strategy Implementation Our core strategy implements simultaneous long and short positions with ATR-based position sizing: Strategy Logic: 1. Input: SL multiplier s, TP multiplier p, with p>s 2. For each entry signal: •Open long position: SL = price −s×ATR, TP = price +p×ATR •Open short position: SL = price +s×ATR, TP = price −p×ATR •Track net profit: Pnet =Plong +Pshort 3. Metric: Hedge success rate = count(Pnet >0) total hedges The strategy tests whether the winning position’s profit can systematically outweigh the losing position’s loss, creating net positive returns despite both positions being opened simultaneously. 4 2.9 Systematic Optimization Approach We employ comprehensive grid search across parameter space, following best practices in systematic trading research [8, 9]. This broad exploration ensures we discover true optima rather than boundary artifacts, addressing concerns about overfitting raised in [10]: •Stop-loss multipliers: 0.1 to 2.0 (20 points) •Take-profit multipliers: 1.1 to 3.0 (20 points) •Total combinations: 400 parameter sets per market regime •Evaluation: 6 markets ×4 seeds ×400 parameters = 9,600 simulations This broad exploration ensures we discover true optima rather than boundary artifacts. 3 Results 3.1 Performance Landscape Analysis Figure 1: Comprehensive optimization analysis across market regimes. Top: Contour plots comparing success rate (left column) and total PnL (right column) for high volatility (top row) and ranging markets (bottom row), revealing distinct optimal parameter regions. Bottom: Scatter plots showing the relationship between profit/loss ratio and success rate, demonstrating how parameter choices affect strategy performance differently across market conditions. The analysis highlights the fundamental tradeoff between success frequency and profit magnitude in systematic hedging strategies. Figure 1 reveals several critical insights: •The success rate optimum is sharp and located at unexpectedly wide stop-loss values •The PnL optimum differs from the success rate optimum, indicating a fundamental tradeoff •Performance surfaces are complex and non-linear, justifying comprehensive exploration 5 3.2 Equity Curve Evolution Figure 2: Equity curves for key parameter sets in high volatility (left) and ranging (right) markets. The optimal parameters (s=1.37, p=1.50) show remarkable consistency but modest absolute returns, while other parameter sets exhibit higher variability with greater profit potential. Figure 2 demonstrates how different parameter choices affect strategy behavior: •Optimal parameters: High consistency, low drawdowns, modest returns •Original ”optimum”: Moderate consistency with better absolute returns •Balanced parameters: Tradeoff between consistency and return magnitude 3.3 Key Parameter Performance Table 2: Performance comparison of key parameter sets (averaged across 4 seeds) Parameter Set Success Rate Avg Profit/Hedge Total PnL Sharpe Ratio Original (0.7, 1.5) 79.4% $0.316 $21.46 1.42 Optimal (1.37, 1.50) 97.2% $0.027 $1.94 0.18 Balanced (0.5, 2.0) 73.0% $0.242 $24.18 1.58 Moderate (1.0, 2.0) 81.5% $0.158 $15.80 1.05 Table 2 reveals the fundamental tradeoff: the highest success rate comes at the cost of reduced profit per trade. This illustrates that ”optimal” depends on the trader’s objective function. 6 3.4 Multi-Seed Statistical Validation Table 3: Performance consistency across random seeds for optimal parameters Seed Success Rate Std. Error 95% CI Lower 95% CI Upper 42 97.1% 0.8% 95.5% 98.7% 101 97.3% 0.7% 95.9% 98.7% 2025 96.8% 0.9% 95.0% 98.6% 1234 97.4% 0.7% 96.0% 98.8% Overall 97.2% 0.3% 96.6% 97.8% Table 3 demonstrates remarkable consistency across random generations, with the 95% confidence interval for success rate being [96.6%, 97.8%]. This statistical robustness validates that the optimal parameters are not artifacts of specific random patterns. 4 Discussion 4.1 The Boundary Optimum Artifact Our research reveals a critical methodological insight: the original ”optimum” at (s=0.70, p=1.50) was a boundary artifact of constrained search space. When we expanded exploration to s∈[0.1,2.0] and p∈[1.1,3.0], the true optimum emerged at (s= 1.37, p = 1.50) with dramatically different characteristics. This finding has profound implications for systematic trading research: •Constrained optimization yields misleading results •True optima may lie in counterintuitive parameter regions •Comprehensive exploration is essential for valid conclusions •Parameter ”reasonableness” should be validated, not assumed 4.2 Risk-Reward Tradeoff Fundamentals The discovered optimal parameters achieve extraordinary consistency (97.2% success rate) but minimal profit per trade ($0.027). This exemplifies the fundamental tradeoff in trading strategy design: Performance = f(Success Rate,Profit per Trade,Transaction Costs) (3) Different parameterizations optimize different aspects of this relationship, and the ”best” parameters depend on the trader’s utility function and risk tolerance. 4.3 Educational Implications This research provides several important lessons for systematic trading education: 4.3.1 Methodological Rigor •Multi-seed validation is essential for reliable conclusions •Broad parameter exploration prevents boundary artifacts •Statistical inference requires proper sampling and uncertainty quantification 7 4.4 The Perils of Misguided Optimization Our research reveals a critical danger in systematic trading development: mathematically correct optimization toward inappropriate objectives. The discovered parameters achieving 97.2% success rate represent a textbook case of this phenomenon: •Technical Perfection, Practical Failure: We successfully optimized for success rate using rigorous methodology, yet produced parameters with negligible profitability •Objective Function Selection: The choice of what to optimize proves as critical as how to optimize •Metric Myopia: Focusing on a single performance dimension (success rate) while ignoring complementary metrics (profit magnitude) creates dangerous blind spots •False Confidence: Statistically significant results can provide unwarranted confidence in practically meaningless outcomes This demonstrates that optimization rigor alone cannot compensate for poor objective selection. Systematic traders must therefore: 1. Define Clear Objectives: Explicitly state what constitutes ”success” beyond mathematical metrics 2. Consider Multiple Dimensions: Evaluate tradeoffs between consistency, profitability, risk, and practical constraints 3. Validate Practical Relevance: Ensure mathematical optima translate to meaningful real-world performance 4. Question Optimization Goals: Regularly re-evaluate whether the chosen objective function aligns with actual trading goals The danger lies not in the optimization methodology, but in applying sophisticated techniques to poorly chosen targets—creating the illusion of scientific rigor while missing practical utility. 4.4.1 Strategy Development •Success rate and profit magnitude represent different optimization objectives •Parameter interactions create complex, non-linear performance surfaces •Market regime significantly impacts optimal parameter choices 4.4.2 Practical Implementation •Hedging strategies can achieve high consistency with proper parameterization •Transaction costs dramatically affect net performance in high-frequency strategies •Real-world implementation requires considering slippage and market impact 5 Conclusion This study demonstrates that systematic parameter optimization reveals strategy behaviors that intuitive approaches miss. The simultaneous long-short hedging strategy, when properly parameterized, can achieve remarkable consistency (97.2% success rate) though with modest absolute returns. Key contributions include: 1. A reproducible framework for systematic strategy optimization 2. Demonstration of boundary optimum artifacts in constrained search spaces 8 3. Multi-seed statistical validation methodology 4. Comprehensive analysis of the success-rate/profit-magnitude tradeoff 5. Critical insight into the dangers of misguided optimization objectives Most importantly, our research sounds a cautionary note: sophisticated optimization techniques applied to inappropriate objectives can produce mathematically impeccable but practically worthless results. The 97.2% success rate optimum stands as a stark reminder that methodological rigor cannot compensate for poor goal selection. Future systematic trading research must therefore prioritize not only how we optimize, but equally what we choose to optimize. The research emphasizes that true strategy understanding requires moving beyond intuitive parameter ranges and embracing comprehensive, statistically rigorous optimization approaches—while maintaining constant vigilance that our optimization objectives align with real-world trading success. References [1] Hull, J. C. (2020). Options, futures and other derivatives (11th ed.). Pearson Education. [2] Tsay, R. S. (2010). Analysis of financial time series (3rd ed.). John Wiley & Sons. [3] Hamilton, J. D. (1994). Time series analysis. Princeton University Press. [4] Guidolin, M., and Pedio, M. (2016). Markov switching models in empirical finance. In Advances in Principal Component Analysis (pp. 1-86). Springer. [5] Uhlenbeck, G. E., and Ornstein, L. S. (1930). On the theory of the Brownian motion. Physical Review, 36(5), 823. [6] Vasicek, O. (1977). An equilibrium characterization of the term structure. Journal of Financial Economics, 5(2), 177-188. [7] Dixit, A. K., and Pindyck, R. S. (1994). Investment under uncertainty. Princeton University Press. [8] Prado, M. L. (2018). Advances in financial machine learning. John Wiley & Sons. [9] Chan, E. P. (2021). Quantitative trading: How to build your own algorithmic trading business (2nd ed.). John Wiley & Sons. [10] Bailey, D. H., and L´opez de Prado, M. (2014). The deflated Sharpe ratio: Correcting for selection bias, backtest overfitting, and non-normality. The Journal of Portfolio Management, 40(5), 94-107. [11] Harvey, C. R., and Liu, Y. (2017). Backtesting. The Journal of Portfolio Management, 40(4), 13-28. 9