Enhancing Trading Strategies Using Model Uncertainty

Most trading systems focus on algorithms for generating entry and exit signals. When the performance deteriorates, developers often try to introduce additional filters and/or modify system parameters.

Reference [1] applied a novel technique, called Dynamic Model Averaging (DMA), to improve model performance. Basically, DMA estimates model uncertainty, and a trade is executed when signals are generated and the model uncertainty is low.

DMA, widely applied in forecasting inflation, S&P 500 volatilities, and exchange rates, dynamically assigns a model probability to each candidate model, enabling time-varying parameters. It aggregates forecasts from all models, using Kalman filtering for estimation and updating model probabilities based on historical forecast accuracy, yielding robust out-of-sample predictions.

The authors pointed out,

We have proposed augmented trading strategies by incorporating considerations of market entry timing. Leveraging estimations from the DMA approach, two criteria are employed to determine optimal market entry times: (1) low uncertainty regarding the model used to forecast trading returns, and (2) positive forecasted trading returns. Subsequently, spanning from April 4th, 2001, to December 31st, 2023, we collect daily data from the Chinese stock market to empirically examine our augmented trading strategies. Utilizing lagged trading excess returns and nine higher-order moments of market performance as market indicators, we forecast future excess returns in both momentum and reversal trading.

Results affirm our augmented strategies yield significant positive returns, surpassing canonical momentum and reversal trading. Canonical strategies mostly saw negative average returns over the period, except 1-day momentum. Conversely, augmented strategies reliably produced positive returns, transaction costs notwithstanding, with most showcasing over 7 % average annual absolute returns. Implementation of our criteria didn’t notably diminish trading chances, selected entry days constituting over 12 % of total. Selected entry days were evenly spread, indicating brief waiting periods for trading.

In short, by applying the DMA approach to estimate model uncertainty and taking signals when the uncertainty is low, the authors managed to greatly improve the performance of momentum and reversal trading strategies.

This is an innovative technique in trading system design. Let us know what you think in the comments below or in the discussion forum.

References

[1] Wenhao Wang, Qingyi Zhang, Pengda An, Feifei Cai, Momentum and reversal strategies with low uncertainty, Finance Research Letters Volume 68, October 2024, 105970

Post Source Here: Enhancing Trading Strategies Using Model Uncertainty

source https://harbourfronts.com/enhancing-trading-strategies-using-model-uncertainty/

Volatility Risk Premium Seasonality Across Calendar Months

Seasonality in investing refers to the tendency of financial markets or specific assets to exhibit predictable patterns at certain times of the year. These patterns can arise due to recurring economic, behavioral, or institutional factors. Understanding and analyzing seasonal trends can help investors time their trades more effectively and enhance portfolio performance.

We have recently discussed the seasonality of the volatility risk premium (VRP) in terms of days of the week. In this regard, Reference [1] examined the VRP in terms of months of the year. The authors pointed out,

As the first in the literature, this study documents a statistically significant December effect, namely, the delta-hedged returns in December are substantially lower than those in other months of the year. The lower hedged returns in December are attributed to overvaluation of options at the beginning of the month, which in turn is attributed to option investors’ consistent failure of recognizing and incorporating the lower realized volatility in the second half of December, i.e., the implied volatility at the beginning of December is consistently larger than the realized volatility in December. This December effect prevails in both equity options and S&P 500 index options. A trading strategy selling straddles based on the decile with the biggest predicted difference between implied volatilities and realized volatilities can generate a monthly return of 13.09% in December, compared with the unconditional sample mean of 0.88%. The next step of the study is to examine and rule out alternative channels such as time-varying risks and demand pressure.

In short, the authors concluded that the VRP is greatest in December and smallest in October.

An explanation for the large VRP in December is that during the holiday season, firms might refrain from releasing material information, leading to low trading volumes. The combination of low trading volume and the absence of important news releases would naturally result in lower realized volatility.

This is another important contribution to the understanding of the VRP. Let us know what you think in the comments below or in the discussion forum.

References

[1] Wei, Jason and Choy, Siu Kai and Zhang, Huiping, December Effect in Option Returns (2025). https://ift.tt/75k1LRn

Article Source Here: Volatility Risk Premium Seasonality Across Calendar Months

source https://harbourfronts.com/volatility-risk-premium-seasonality-across-calendar-months/

Extension of the Black-Scholes-Merton Model to Include Supply Change Rate for Ethereum Options

Ethereum (ETH) is a cryptocurrency that resembles a combination of a currency, a stock, and a commodity. It is a non-dividend-paying crypto asset with a dynamic supply change parameter. Ethereum options have been traded since 2019. Deribit is the largest ETH options exchange by volume, with a market share of approximately 80%.

A particularity of ETH is the changing nature of its supply. Specifically, the supply change rate can be expressed as follows,

Supply change rate = Net issuance rate = Issuance rate − Burn rate

Reference [1] generalized the Black-Scholes-Merton (BSM) formalism to include ETH’s supply change rate. The authors pointed out,

The net issuance rate introduces skewness in volatility structures. This skewness is weighted for deep-in-the-money options, which is consistent with the behavior of option prices. As strike prices increase significantly, the implied volatilities asymptotically approach each other.

Moreover, the sensitivity of the results to even minor changes in the net issuance rate parameter is noteworthy. Figures 12 - 13 demonstrate this effect. This indicates that the inclusion of the net issuance rate can cause significant changes in option pricing. Consequently, options could be fundamentally mispriced if the parameter is completely ignored.

The main contribution of the thesis is the identification of a deterministic factor in the pricing of crypto asset options, the supply change rate, which is not taken into account in the traditional BSM model. The extended BSM model, or alternatively the crypto asset BSM, presented in this thesis includes this rate in the model. The supply change rate can take both positive and negative values within its mathematically defined limits. Moreover, the crypto asset BSM can be used for any other crypto asset that has a supply change parameter, preferably with low block times.

Basically, the author employed the formula used for incorporating stock dilution effects and extended it to ETH options.

Another interesting insight from the paper is that it shows the volatility smirk of ETH options, where, unlike equity options, out-of-the-money call options have higher implied volatility than at-the-money calls.

Let us know what you think in the comments below or in the discussion forum.

References

[1] Teemu Laurikainen, An extension of the Black-Scholes-Merton options pricing model to Ethereum, Aalto University, 2025

Post Source Here: Extension of the Black-Scholes-Merton Model to Include Supply Change Rate for Ethereum Options

source https://harbourfronts.com/extension-black-scholes-merton-model-include-supply-change-rate-ethereum-options/

Time Series vs. Machine Learning: A Systematic Evaluation

Forecasting is important in finance, as it helps investors, analysts, and institutions make informed decisions under uncertainty. Up to now, most forecasting techniques have relied on traditional time series methods, such as ARIMA, GARCH, and exponential smoothing. However, with recent advancements in machine learning and artificial intelligence, these technologies have increasingly found applications in financial forecasting. Their ability to capture complex, nonlinear relationships and process large volumes of data has opened new possibilities for improving prediction accuracy in areas such as asset prices, volatility, and risk.

Reference [1] presents a systematic comparison of traditional time series techniques with newer AI/ML approaches. It highlights the weaknesses of traditional time series methods, notably they assume stationarity and linear relationships, which often do not hold in financial markets. These models struggle with non-stationary data, non-linear dynamics, and large datasets, limiting their ability to capture the full complexity of market behavior.

The paper also discusses the advantages of AI-driven methods, particularly that they excel at capturing complex, non-linear relationships in financial data, and adapting to changing market conditions without manual intervention. They also handle large, high-dimensional datasets effectively, uncovering hidden patterns and making more accurate predictions than traditional models.

The authors made several comparisons using criteria such as,

• Accuracy
• Computational Complexity
• Flexibility and Adaptability
• Interpretability

The authors pointed out,

The comparison of both approaches revealed that while traditional methods are more interpretable and computationally efficient, AI-driven techniques provide greater accuracy and adaptability, especially when dealing with the dynamic and volatile nature of modern financial markets. However, the challenge of obtaining high-quality, reliable data and avoiding overfitting remains for both types of models.

In practice, the decision to use traditional methods versus AI-driven approaches depends largely on the nature of the financial data and the specific forecasting needs. Traditional methods may still be the preferred choice for simpler, well-behaved datasets where linearity and stationarity are present, or when computational resources are limited. They are also suitable for scenarios where interpretability is essential, such as regulatory environments or when model transparency is required. Conversely, AI-driven models should be considered when forecasting complex, non-linear, or high-dimensional financial data, such as stock prices or forex rates, where traditional models struggle. These models are particularly useful when predictive accuracy is paramount, and sufficient computational resources are available to handle the increased complexity.

In short, the new AI/ML techniques offer advantages but also come with disadvantages. However, nothing prevents us from combining these two approaches and leveraging their respective strengths.

Let us know what you think in the comments below or in the discussion forum.

References

[1] Gwokkwan Sun, and Shuhan Deng, Financial Time Series Forecasting: A Comparison Between Traditional Methods and AI-Driven Techniques, Journal of Computer, Signal, and System Research, Vol. 2 No. 2 (2025)

Post Source Here: Time Series vs. Machine Learning: A Systematic Evaluation

source https://harbourfronts.com/time-series-vs-machine-learning-systematic-evaluation/

Tail Risk Hedging with Corporate Bond ETFs

Tail risk hedging is a strategy designed to protect portfolios against extreme market moves that occur infrequently but have a significant impact when they do. These “tail events” lie at the far ends of a return distribution and often coincide with financial crises, sharp market crashes, or systemic shocks. A well-structured tail risk hedge, typically involving options or volatility instruments, can provide substantial value during periods of heightened uncertainty.

Reference [1] proposed a tail risk hedging scheme by shorting corporate bonds. Specifically, it constructed three signals—Momentum, Liquidity, and Credit—that can be used in combination to signal entries and exits into short high-yield ETF positions to hedge a bond portfolio.  The authors pointed out,

The research above constructed signals on the Investment Grade bond market to inform a dynamic hedge that deploys liquid bond ETFs as hedges to effectively and quickly protect high carry bond funds. It succeeded in lowering absolute and relative risk, increasing annualised returns, and improving Sortino for PIMIX and avoiding drawdowns for DODIX, in a realistic framework that incorporates trading costs, funding costs, and volume sized hedge positions.

Credit Risk, Liquidity, and Momentum signals derived from options, duration times spread, and cumulative duration-neutral returns respectively, each seemed to capture some orthogonal information about the IG bond market. Hedge performance considering individual signals, followed by their combination, proves this point - with an optimal improvement in Sortino of ≥ 0.7 using the joint signals. When searching the hedge model’s parameter space, results remain strong and consistent over a wide array of tested parameters.

Hedging is cost effective as the research has focused on establishing short positions in IG (LQD) and HY (HYG) bond ETFs rather than shorting individual IG corporate bonds. IG bond ETFs are liquid and have low bid-ask spreads, and establishing shorts in the IG bond ETF space via LQD & HYG provides great downside convexity which benefits the efficacy of the hedge. While IG and HY CDXs have far larger traded volumes than LQD & HYG, they do not have the same downside convexity and prove to be not as effective as ETFs

In short, it's possible to develop an effective tail risk hedging strategy using corporate bond ETFs.

An interesting insight from this paper is that it points out how using corporate ETFs benefits from downside convexity while using credit default swaps such as IG CDXs does not.

Let us know what you think in the comments below or in the discussion forum.

References

[1] Travis Cable, Amir Mani, Wei Qi, Georgios Sotiropoulos and Yiyuan Xiong, On the Efficacy of Shorting Corporate Bonds as a Tail Risk Hedging Solution, arXiv:2504.06289

Originally Published Here: Tail Risk Hedging with Corporate Bond ETFs

source https://harbourfronts.com/tail-risk-hedging-corporate-bond-etfs/

Machine Learning for Algorithmic Trading: A Comprehensive Review

Thanks to the advancement in computing technologies, we’re seeing more widespread use of machine learning, especially deep learning, in the financial services sector. It’s no longer just a theoretical tool; it's showing up in everything from credit risk models to algorithmic trading strategies.

Reference [1] provides a comprehensive review of deep learning techniques used in the financial sector, with a focus on algorithmic trading. It offers a structured analysis of deep learning’s applications across various areas of trading, aiming to identify key trends, challenges, and emerging opportunities by critically evaluating existing research.

The paper provides detailed insights into methodologies applied in different sub-areas of trading such as,

• Stock price prediction
• Market volatility prediction
• Portfolio optimization
• Sentiment analysis for trading
• Risk management
• Anomaly detection and fraud detection
• Supply chain forecasting

Specifically, in volatility forecasting, it highlights,

Recent studies have emphasized the significance of incorporating multiple data streams, including macroeconomic indicators, sentiment analysis, and high-frequency trading data, to enhance the predictive accuracy of volatility models [129,130]. The findings suggest that hybrid models outperform single-model approaches, but data noise and overfitting remain challenges. As shown in Table 8, a variety of models have been applied to different datasets, each with specific contributions and limitations.

Overall, the authors concluded,

This review has highlighted the transformative potential of deep learning in algorithmic trading, where models such as LSTM, CNN, and Reinforcement Learning have shown substantial improvements in predicting financial markets and optimizing trading strategies. However, significant challenges remain, particularly related to data quality, overfitting, and the interpretability of complex DL models. Financial markets are noisy, volatile, and influenced by a multitude of factors, making it difficult for models to generalize well. Additionally, the black-box nature of DL models raises concerns for traders and regulators who require transparency in decision-making. Emerging trends such as attention mechanisms, transformer architectures, and hybrid models offer promising solutions to these challenges, alongside integrating alternative data sources like social media sentiment and news. Future research must focus on improving model robustness, developing explainable AI techniques, and addressing computational efficiency to unlock the full potential of DL in real-world trading environments. By overcoming these hurdles, DL can significantly enhance the accuracy and effectiveness of algorithmic trading, providing traders with more powerful tools for navigating complex financial markets.

In short, deep learning is useful but still has its limitations.

In our experience, being able to leverage advances in computing is definitely an edge, but domain knowledge remains essential.

Let us know what you think in the comments below or in the discussion forum.

References

[1] MD Shahriar Mahmud Bhuiyan, MD AL Rafi, Gourab Nicholas Rodrigues, MD Nazmul Hossain Mir, Adit Ishraq, M.F. Mridha, Jungpil Shin, Deep learning for algorithmic trading: A systematic review of predictive models and optimization strategies, Array, Volume 26, 2025, 100390,

Post Source Here: Machine Learning for Algorithmic Trading: A Comprehensive Review

source https://harbourfronts.com/machine-learning-algorithmic-trading-comprehensive-review/