# Regime-Switching Hidden Markov Models for Adaptive Swing Strategies

## The Non-Stationarity of Financial Markets

Financial markets operate as fundamentally non-stationary ecosystems. They alternate between behavioral phases characterized by distinct return distributions, correlation structures, and volatility profiles [cite: 1]. Standard swing trading strategies, which seek to capture multi-day to multi-week price movements, historically rely on static parameters optimized over long-term averages. These static models assume that market data is drawn from a single, continuous probability distribution. In reality, markets switch abruptly between periods of calm trending behavior and chaotic, high-volatility mean-reversion [cite: 2, 3, 4]. 

When a structural macro shift occurs, static strategies frequently suffer severe drawdowns because their risk management and alpha generation logic remain anchored to the preceding environment [cite: 2, 3]. The assumption of a stationary covariance matrix is a known systemic vulnerability. For example, during the 2022 global tightening cycle, the historical negative correlation between equities and bonds shifted to a positive correlation, severely impairing risk parity strategies that relied on static covariance estimations [cite: 5]. Top-tier quantitative institutions, such as Two Sigma and AQR Capital Management, actively condition their covariance matrices on the current market state to avoid systematically underestimating factor correlations during crisis periods [cite: 5].

To solve the limitations of static parameters, systematic hedge funds utilize regime-switching frameworks to perform market state detection. By analyzing the statistical fingerprints left by price action, return dispersion, and volatility clustering, adaptive swing trading systems dynamically modulate position sizing, factor exposures, and trailing stop-loss distances [cite: 4, 6]. 

## Regime Detection Methodologies

The core objective of market regime detection is to infer an unobservable environmental state from noisy, observable financial data. Researchers and quantitative analysts generally rely on three primary statistical frameworks: Gaussian Mixture Models (GMM), Bayesian Change-Point Detection (BCPD), and Hidden Markov Models (HMM) [cite: 7, 8, 9]. 

The mathematical framework selected dictates how a trading system interprets temporal dependencies in the data series. 

### Comparison of Regime Detection Frameworks

The following table summarizes the structural differences, temporal assumptions, and practical trading applications of the three dominant regime detection methodologies.

| Feature | Gaussian Mixture Models (GMM) | Bayesian Change-Point Detection (BCPD) | Hidden Markov Models (HMM) |
| :--- | :--- | :--- | :--- |
| **Core Mechanism** | Unsupervised clustering of data points into multiple Gaussian distributions based on value proximity [cite: 8]. | Calculates the posterior probability of a structural break at every time step to isolate homogeneous segments [cite: 7, 9]. | Models a system as a sequence of discrete latent states with distinct emission and transition probabilities [cite: 4, 10]. |
| **Temporal Dependency** | Independent and identically distributed (i.i.d.) assumption. Ignores the chronological sequence of the data [cite: 8, 11]. | Assumes a sequence of homogeneous time segments separated by abrupt shifts or policy interventions [cite: 9, 12]. | Markovian property: The probability of the current state depends entirely on the immediately preceding state [cite: 10]. |
| **Primary Strength** | Flexible approximation of arbitrary, unknown probability distributions without assuming time structure [cite: 8]. | Provides exact probabilistic localization of sudden macroeconomic shifts, highly useful for post-event analysis [cite: 9]. | Excels at capturing volatility clustering and the highly persistent nature of financial market phases [cite: 4, 13]. |
| **Primary Weakness** | Inability to model the persistence of a market state over time, leading to rapid, noisy classifications in live trading [cite: 8, 11]. | Highly sensitive to prior parameter calibration; struggles to detect complex variance shifts independently of mean shifts [cite: 9]. | Computationally demanding for large sequences; risk of overfitting if the number of discrete states is set too high [cite: 14, 15, 16]. |

Among these, Hidden Markov Models are overwhelmingly preferred for the development of adaptive swing trading strategies because financial market regimes exhibit extreme temporal persistence [cite: 13]. The tendency of a high-volatility bear market to remain highly volatile for several weeks aligns perfectly with the Markovian assumption of state transitions, enabling traders to deploy capital with an understanding of the immediate probabilistic environment [cite: 1, 13].

### Mechanics of Hidden Markov Models

A continuous Gaussian HMM applied to financial time series is defined by a specific set of parameters, denoted as $\lambda = \{A, \mu, \sigma, \pi\}$ [cite: 10]. 

*   **$A$ (Transition Matrix):** The square matrix defining the probability of transitioning from one hidden state to any other hidden state.
*   **$\mu$ and $\sigma$ (Emission Probabilities):** The mean and variance (or covariance matrix in multivariate models) of the observable market data generated by each hidden state.
*   **$\pi$ (Initial Probabilities):** The vector of probabilities indicating the starting state of the sequence [cite: 10].

The raw inputs for an HMM must be engineered for stationarity. Feeding raw asset prices directly into the model results in spurious and unstable classifications. Standard practice involves transforming prices into stationary features, such as log returns, short-term moving average residuals, or rolling variance metrics [cite: 4, 10]. For example, a three-state HMM designed to trade the S&P 500 index relies heavily on daily percentage returns and a rolling 10-day mean squared error to approximate volatility [cite: 10]. 

### Regime Classification and Economic Profiles

When applied to broad equity indices, an HMM utilizing unsupervised learning generally clusters the stationary observations into three intuitive economic profiles [cite: 10, 16]:

1.  **Regime 0 (Sideways/Moderate):** Characterized by near-zero mean returns with moderate volatility. This regime often exhibits high mean-reversion and choppy price action.
2.  **Regime 1 (Bull/Low Risk):** Characterized by positive mean returns with tight, low volatility. This regime supports aggressive trend-following and long-biased factor exposure.
3.  **Regime 2 (Bear/Crisis):** Characterized by negative mean returns with extreme, erratic volatility. This regime features violent intraday swings and requires defensive positioning or short-volatility reduction [cite: 10, 16].

Formal information criteria, such as the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), sometimes suggest that financial data supports five or six distinct states [cite: 16]. However, quantitative researchers typically constrain the model to three or four states to prevent overfitting and to ensure the regimes remain economically interpretable and actionable for trade execution [cite: 15, 16]. 

### Transition Probability Matrices and Regime Persistence

The transition probability matrix ($A$) is the predictive engine of the HMM. It quantifies the likelihood of the market remaining in its current state versus shifting to a new one. Empirical studies consistently demonstrate extreme diagonal dominance in financial transition matrices. This confirms that the probability of remaining in the current state is overwhelmingly higher than the probability of switching, a phenomenon known as regime persistence [cite: 4, 17].



Research on aggregated systemic risk indices highlights significant asymmetric transition dynamics. Financial markets rarely jump directly from a stable, low-risk bull state into a high-volatility crisis state. Instead, the moderate or sideways regime serves as a transitional gateway [cite: 18, 19].

[image delta #1, 0 bytes]

 Recognizing this structural gateway provides systematic traders with a critical lead time—often averaging up to 30 days—to de-risk portfolios before a full crisis regime materializes [cite: 18].

When deployed in live trading environments, the HMM relies on two primary algorithms. The Baum-Welch algorithm, an iterative expectation-maximization method, optimizes the model parameters ($\mu, \sigma, A$) based on historical training data [cite: 4, 10]. Once the parameters are fixed, the Viterbi algorithm recursively decodes the live observation sequence to infer the most probable hidden state sequence, providing the algorithm with a real-time actionable regime signal [cite: 4, 10, 20]. 

## Advanced Hidden Markov Model Architectures

While standard continuous Gaussian HMMs offer robust baseline capabilities, they carry structural limitations. Chiefly, they assume that transition probabilities are strictly time-invariant. In reality, the fundamental rules governing how markets switch states can evolve during major macroeconomic events, geopolitical shocks, or monetary policy shifts. To address these limitations, quantitative researchers integrate complex architectural modifications.

### Adaptive Hierarchical Hidden Markov Models

An Adaptive Hierarchical Hidden Markov Model (AH-HMM) separates fast-moving daily or weekly market states from a slower, overarching structural layer. It introduces an unobserved meta-regime layer that reflects the broader macro-financial environment, such as periods of high or low systemic uncertainty [cite: 21]. 

Under the AH-HMM framework, each meta-regime possesses its own distinct transition matrix. The model is capable of "structural learning," recognizing that the mechanics of state switching are fundamentally altered by the broader environment. Empirical applications to indices like the S&P 500 and EURO STOXX 50 indicate that turbulent and bear market regimes become significantly more persistent during high-uncertainty macro environments [cite: 21]. The hierarchical structure allows the AH-HMM to attain higher in-sample likelihoods and more balanced Value-at-Risk (VaR) coverage than conventional, static-transition HMMs [cite: 21].

### Statistical Jump Models and Jump-Diffusion

A secondary limitation of the standard HMM is its struggle to simultaneously reproduce heavy-tailed return distributions and persistent volatility clustering. Standard jump-diffusion models are effective at capturing sudden, discrete large-magnitude price movements generated by earnings surprises or macroeconomic announcements [cite: 22]. However, post-jump volatility in standard diffusion models immediately reverts to its baseline level, failing to capture the empirically observed persistence of crisis volatility [cite: 22].

Hybrid HMM frameworks solve this by merging discrete market states with a Poisson-driven jump-duration mechanism [cite: 22, 23]. This architectural adjustment forces the model to dwell in high-volatility tail states for empirically realistic durations, connecting gradual price evolution with sudden large moves [cite: 22]. The implementation of Statistical Jump Models (JM), which apply a jump penalty at each state transition to artificially enforce regime stability, consistently outperforms traditional Markov-switching models in out-of-sample backtests, resulting in enhanced risk-adjusted returns and reduced strategy turnover [cite: 24, 25].

### Deep Learning Integrations

The integration of deep neural networks with HMMs represents the frontier of regime detection. In dual-model alpha generation systems, the HMM is utilized strictly to capture temporal dependencies and broad market regimes, while a fully connected feedforward neural network or Long Short-Term Memory (LSTM) network is tasked with learning subtle, non-linear price patterns within that specific regime [cite: 26]. 

While deep learning models like LSTMs and Transformers excel at addressing vanishing gradients and sequence prediction, they suffer from a "black-box" nature that makes risk management opaque. Hybridizing them with HMMs allows the system to ground the neural network's predictions within a statistically interpretable market state, blending the pattern recognition power of AI with the probabilistic rigor of state-space modeling [cite: 14, 26].

## Designing Regime-Adaptive Swing Strategies

The output of any Hidden Markov Model is fundamentally diagnostic; it identifies the probabilistic environment but does not execute trades [cite: 2, 6]. To build a functional adaptive swing trading strategy, the inferred state probabilities must be programmatically mapped to specific execution parameters.

[image delta #2, 0 bytes]

 Systematic managers utilize these regime signals to dictate asset allocation, risk budgeting, position sizing, and stop-loss calibration [cite: 27, 28, 29, 30].



### Volatility-Scaled Position Sizing

A prevalent architectural flaw in retail swing trading is the conflation of position size with stop-loss distance. In institutional systematic trading, these are independent variables treated sequentially. Position sizing is dictated exclusively by the amount of risk (capital) the algorithm allocates to a specific market, irrespective of the trading speed or the precise placement of the stop-loss order [cite: 28]. 

Because an asset's standard deviation varies radically by regime—sometimes tripling during a macroeconomic shock—static position sizes result in highly erratic portfolio risk [cite: 3]. A $100,000 equity position in an HMM-classified low-volatility regime carries dramatically less daily capital-at-risk than a $100,000 position in a high-volatility crisis regime. 

Adaptive systems resolve this imbalance via volatility scaling. The nominal position size is scaled inversely to the asset's current volatility, effectively allocating a fixed number of "risk units" rather than a fixed dollar amount [cite: 30, 31, 32]. Large managed futures firms, such as Man Group's AHL, heavily utilize this methodology across hundreds of global markets [cite: 31, 32]. When the HMM detects a transition from a calm regime to a highly volatile bear regime, the system automatically cuts the nominal position size to maintain constant risk parity [cite: 4]. This dynamic adjustment prevents the strategy from being mathematically overrun by the expanded standard deviation of asset returns, effectively smoothing the equity curve regardless of the underlying market chaos [cite: 3].

### Dynamic Stop-Loss Calibration

While position size governs the overall portfolio risk budget, stop-loss distances manage the speed of trading and the strategy's tolerance for intraday market noise [cite: 28]. In a fast, volatile market, trailing stops must be set wider to avoid premature liquidation from random price fluctuations. Conversely, in slow, trending markets, stops can be tightened to securely lock in accumulated profits [cite: 28].

The industry standard for measuring this noise is the Average True Range (ATR), which quantifies the absolute average price movement over a specified lookback window. A baseline parameter for momentum swing trading often relies on a 14-period ATR with a 2.0x multiplier. This places the stop-loss at a distance of twice the average daily range away from the entry price or recent swing high [cite: 33, 34].

In a regime-adaptive framework, the ATR multiplier itself becomes a dynamic variable conditioned directly on the HMM output state. 

*   **Low-Volatility Trend Regimes:** When the HMM signals stable persistence, the system tightens the trailing stop multiplier (e.g., to 1.5x ATR) to protect capital against sudden mean-reversion. Because trends are well-defined and intraday noise is minimal, a tighter stop effectively captures the bulk of the move [cite: 35, 36].
*   **High-Volatility Bear Regimes:** During structural downturns, market swings are violent, and short-term mean reversion is highly prevalent. To avoid stop-hunting and premature exits caused by liquidity vacuums, the system expands the ATR multiplier (e.g., to 3.0x ATR). This expansion provides the trade with the necessary "breathing room" to survive intraday whipsaws, provided that the overall nominal position size has already been reduced via the volatility scaling step [cite: 34, 36]. 

Recent applications in cryptocurrency swing trading—such as the AdaptiveTrend framework—utilize 6-hour intervals with dynamic trailing stops calibrated explicitly to intraday volatility regimes. By dynamically adjusting the ATR multiplier based on the latent market state, this specific framework achieved annualized Sharpe ratios in excess of 2.40, significantly outperforming static time-series momentum models [cite: 37].

### Factor Rotation and the Information Ratio

Beyond managing downside risk, Hidden Markov Models dictate strategy selection and alpha generation. The Information Ratio (IR)—which measures benchmark-beating active return per unit of active risk—can collapse rapidly during regime shifts. This collapse often occurs not because the underlying alpha signal is fundamentally broken, but because the volatility of the Information Coefficient (IC) spikes drastically [cite: 38]. 

The Fundamental Law of Active Management dictates that long-only constraints destroy Information Ratios during high-volatility events by preventing the full expression of negative alpha signals and eliminating the ability to dynamically hedge [cite: 38]. To counteract this, quantitative managers utilize regime-switching models to rotate capital between distinct, uncorrelated factor strategies [cite: 10, 39]. 

For instance, a systematic strategy might maintain an aggressive, leveraged posture on value and momentum factors during a confirmed HMM "Bull" regime [cite: 10]. If the HMM detects a structural shift to a "High Volatility Chop" regime, the strategy dynamically rotates capital out of directional momentum and into market-neutral models, short-duration defensive assets, or quality-factor baseline portfolios [cite: 10, 35]. This structural flexibility ensures that the portfolio is constantly trading the optimal factor for the current environment, bypassing the rigid mandate constraints that typically destroy returns during macroeconomic shocks [cite: 38].

## Empirical Performance Metrics and Risk Mitigation

The economic value of regime-switching strategies cannot be measured merely by absolute return. Evaluating a systematic strategy solely by its long-term, static Sharpe ratio is dangerously misleading, as it assumes that all historical returns were drawn from a single, continuous distribution [cite: 3]. 

### Regime-Conditional Performance Decomposition

Rigorous backtesting of an adaptive strategy requires a regime-conditional performance decomposition [cite: 3, 37]. Analysts must evaluate the strategy's Sharpe ratio, maximum drawdown, and win rate strictly within the segmented boundaries of each HMM-identified regime [cite: 3].

Empirical testing reveals stark divergences across market states. A short-volatility strategy, for example, may exhibit a highly favorable Sharpe ratio during a multi-year low-volatility trend, but suffer catastrophic, account-clearing drawdowns in an HMM-flagged crisis regime [cite: 3]. The February 2018 volatility spike ("Volmageddon") serves as a prime example, where short-volatility products with years of flawless track records were mathematically wiped out in a single trading session because their models failed to detect and adapt to a structural regime transition [cite: 3]. 

By filtering trades through an HMM—for example, explicitly prohibiting trend-following long trades when the probability of a sideways or bear regime exceeds a defined confidence threshold (e.g., 80%)—the system systematically truncates the left tail of the return distribution [cite: 35, 40, 41].

### Drawdown Mitigation and Strategy Resilience

The integration of Hidden Markov Models into asset allocation and swing trading routinely yields superior risk-adjusted metrics compared to static benchmarks. In a comprehensive backtest of the S&P 500 spanning over a decade, a regime-switching factor model guided by a continuous Gaussian HMM achieved a Sharpe ratio of 2.017 and an Information Ratio of 1.64. This performance significantly outpaced individual, static factor models (such as the Carhart four-factor or Fama-French modified variants) [cite: 10, 39]. The HMM-guided model successfully preserved capital during severe market downturns by rotating into defensive postures, resulting in a highly restrained maximum drawdown of just 12.83% [cite: 10].

Similarly, out-of-sample testing of Statistical Jump Models across US, German, and Japanese equity indices confirms that regime-guided strategies consistently reduce aggregate volatility and maximum drawdown while enhancing the overall Sharpe ratio against standard buy-and-hold methodologies [cite: 24, 25].

## Institutional Implementations and Future Directions

The deployment of these models at scale by firms like Man Group, AQR, and Two Sigma demonstrates their viability, but retail and mid-tier institutional practitioners face distinct implementation hurdles [cite: 5, 31, 42]. 

While theoretical HMM models show immense promise, their live performance is highly sensitive to operational friction. Transaction costs, bid-ask spread crossing, and execution delays can erode the alpha generated by regime switching, particularly if the HMM is configured with too many hidden states, resulting in excessive portfolio turnover [cite: 10, 24, 40]. The calibration of the expectation-maximization threshold is critical; setting the iteration limit correctly is necessary to balance regime granularity and prevent the model from overfitting to historical noise [cite: 15].

Furthermore, as traditional asset classes become increasingly integrated, the reliance on purely price-derived features for HMM emission probabilities is evolving. Future research is directing attention toward the incorporation of alternative data, natural language processing sentiment scores, and macroeconomic leading indicators directly into the multivariate feature space of the HMM [cite: 43, 44, 45]. By augmenting the HMM with macroeconomic inputs like yield curve spreads, inflation data, and commodity spot prices, predictive models can achieve a more holistic understanding of the fundamental drivers behind regime transitions [cite: 44].

Ultimately, the application of Hidden Markov Models transforms swing trading from a rigid, rules-based endeavor into a dynamic, probabilistic science. By mathematically decoupling position sizing from stop-loss logic, and conditioning both strictly on the latent market regime, quantitative systems can systematically harvest alpha while aggressively managing the volatile reality of financial time series.

## Sources

1. [theideafarm.com](https://theideafarm.com/category/portfolio-management/)
2. [toptradersunplugged.com](https://www.toptradersunplugged.com/podcast/profiting-from-different-investment-regimes-ft-rob-carver/)
3. [medium.com](https://medium.com/@navnoorbawa/the-information-ratio-how-aqr-millennium-and-worldquant-built-350b-beating-every-constraint-0e031e7a0989)
4. [pm-research.com](https://www.pm-research.com/content/iijpormgmt/52/2/local/complete-issue.pdf)
5. [pm-research.com](https://www.pm-research.com/content/iijpormgmt/51/3/local/complete-issue.pdf)
6. [iaeng.org](https://www.iaeng.org/IJAM/issues_v55/issue_12/IJAM_55_12_26.pdf)
7. [researchgate.net](https://www.researchgate.net/publication/228315949_Regime_Changes_and_Financial_Markets)
8. [arxiv.org](https://arxiv.org/html/2602.03874v2)
9. [pmc.ncbi.nlm.nih.gov](https://pmc.ncbi.nlm.nih.gov/articles/PMC12419650/)
10. [pm-research.com](https://www.pm-research.com/content/iijjfds/6/4/local/complete-issue.pdf)
11. [arxiv.org](https://arxiv.org/abs/2603.10202)
12. [arxiv.org](https://arxiv.org/html/2603.10202v1)
13. [mdpi.com](https://www.mdpi.com/1911-8074/13/12/311)
14. [quantifiedstrategies.com](https://www.quantifiedstrategies.com/hidden-markov-model-market-regimes-how-hmm-detects-market-regimes-in-trading-strategies/)
15. [cube.exchange](https://www.cube.exchange/what-is/market-regime-detection-with-hidden-markov-models)
16. [backend.orbit.dtu.dk](https://backend.orbit.dtu.dk/ws/files/152802375/phd465_Nystrup_P.pdf)
17. [mdpi.com](https://www.mdpi.com/2227-7390/13/10/1577)
18. [researchgate.net](https://www.researchgate.net/publication/396244436_Change_Point_Detection_in_Financial_Market_Using_Topological_Data_Analysis)
19. [mlg.eng.cam.ac.uk](https://mlg.eng.cam.ac.uk/pub/pdf/Tur11.pdf)
20. [researchgate.net](https://www.researchgate.net/publication/215466378_A_Bayesian_approach_to_change_points_detection_in_time_series)
21. [mdpi.com](https://www.mdpi.com/1911-8074/19/1/15)
22. [mdpi.com](https://www.mdpi.com/1911-8074/13/12/311)
23. [arxiv.org](https://arxiv.org/html/2407.19858v6)
24. [mql5.com](https://www.mql5.com/en/code/mt5/indicators)
25. [osfi-bsif.gc.ca](https://www.osfi-bsif.gc.ca/en/guidance/guidance-library/capital-adequacy-requirements-car-2026-chapter-7-settlement-counterparty-risk)
26. [federalregister.gov](https://www.federalregister.gov/documents/2013/10/11/2013-21653/regulatory-capital-rules-regulatory-capital-implementation-of-basel-iii-capital-adequacy-transition)
27. [bis.org](https://www.bis.org/publ/bcbs265.pdf)
28. [meti.go.jp](https://www.meti.go.jp/meti_lib/report/2021FY/000657.pdf)
29. [reddit.com](https://www.reddit.com/r/mathematics/comments/kcv1ay/hmm_vs_gmm/)
30. [arxiv.org](https://arxiv.org/pdf/2504.21046)
31. [pdfs.semanticscholar.org](https://pdfs.semanticscholar.org/5b52/40059003ac4d8fb186bcbf3aaa4ccb73bd5b.pdf)
32. [mdpi.com](https://www.mdpi.com/2227-7390/13/15/2505)
33. [researchgate.net](https://www.researchgate.net/figure/Comparing-HMM-GMM-HMM-and-LSTM-HMM-with-monthly-or-quarterly-input-and-4-year-6-year_tbl15_361386537)
34. [quantifiedstrategies.com](https://www.quantifiedstrategies.com/hidden-markov-model-market-regimes-how-hmm-detects-market-regimes-in-trading-strategies/)
35. [luxalgo.com](https://www.luxalgo.com/library/indicator/hidden-markov-model-market-regimes/)
36. [researchgate.net](https://www.researchgate.net/publication/347401160_Regime-Switching_Factor_Investing_with_Hidden_Markov_Models)
37. [aimspress.com](https://www.aimspress.com/article/doi/10.3934/DSFE.2025019?viewType=HTML)
38. [mdpi.com](https://www.mdpi.com/1911-8074/13/12/311)
39. [medium.com](https://medium.com/@matthewwang_91639/algorithmic-factor-investing-with-market-regime-classification-6bc2f8c7168b)
40. [jornadaseconometria2018.files.wordpress.com](https://jornadaseconometria2018.files.wordpress.com/2018/09/measuring-and-trading-volatility-a-regime-switching-approach.pdf)
41. [econstor.eu](https://www.econstor.eu/bitstream/10419/203800/1/1032219653.pdf)
42. [hedgenordic.com](https://hedgenordic.com/wp-content/uploads/2020/06/systematic_strategies.pdf)
43. [aima.org](https://www.aima.org/static/f14f296d-3751-45ec-89cfb3120b30ce20/AIMA-Journal-Edition-127.pdf)
44. [man.com](https://www.man.com/documents/download/06abe-b2bb5-bae24-115b4/Man_Funds_VI_plc_-_Man_GLG_High_Yield_Opportunities_Prospectus_incl_Addendum_English_%28Hong_Kong%29_26-04-2024.pdf)
45. [sec.gov](https://www.sec.gov/Archives/edgar/data/1052354/000119312518102087/d443456d10k.htm)
46. [pm-research.com](https://www.pm-research.com/content/iijpormgmt/51/8/local/complete-issue.pdf)
47. [docs.qbea.org](https://docs.qbea.org/)
48. [arc.cct.ie](https://arc.cct.ie/cgi/viewcontent.cgi?article=1122&context=ict)
49. [arxiv.org](https://www.arxiv.org/list/cs/new?skip=25&show=1000)
50. [ideas.repec.org](https://ideas.repec.org/e/c/pch151.html)
51. [turtletrader.com](https://www.turtletrader.com/man-trend-following-fund/)
52. [man.com](https://www.man.com/insights/lss-trendsetters-mechanics-of-trend-following)
53. [man.com](https://www.man.com/insights/skewness)
54. [man.com](https://www.man.com/insights/volatility-is-back-better-to-target-returns-or-target-risk)
55. [man.com](https://www.man.com/insights/deep-dive-trend-following)
56. [google.com](https://www.google.com/search?q=time+in+Toronto,+CA)
57. [mdpi.com](https://www.mdpi.com/1911-8074/13/12/311)
58. [ideas.repec.org](https://ideas.repec.org/p/arx/papers/2402.05272.html)
59. [arxiv.org](https://arxiv.org/abs/2402.05272)
60. [regimeforecast.com](https://regimeforecast.com/blog/regime-dependent-strategy-performance)
61. [politesi.polimi.it](https://www.politesi.polimi.it/retrieve/a4fb0146-1c24-43c2-bc13-dc26f729a2a8/2025_12_Jia_executive_summary_02.pdf)
62. [theideafarm.com](https://theideafarm.com/category/portfolio-management/)
63. [medium.com](https://medium.com/@navnoorbawa/the-information-ratio-how-aqr-millennium-and-worldquant-built-350b-beating-every-constraint-0e031e7a0989)
64. [evanstoncap.com](https://www.evanstoncap.com/uploads/downloads/Evanston-Capital_2024-Hedge-Fund-Outlook.pdf)
65. [arxiv.org](https://arxiv.org/pdf/2006.05632)
66. [scribd.com](https://www.scribd.com/document/330897660/AQR-Portfolio-Rebalancing-Part-1-Strategic-Asset-Allocation)
67. [quantifiedstrategies.com](https://www.quantifiedstrategies.com/hidden-markov-model-market-regimes-how-hmm-detects-market-regimes-in-trading-strategies/)
68. [luxalgo.com](https://www.luxalgo.com/library/indicator/hidden-markov-model-market-regimes/)
69. [researchgate.net](https://www.researchgate.net/publication/215466378_A_Bayesian_approach_to_change_points_detection_in_time_series)
70. [scribd.com](https://www.scribd.com/document/1025298368/Search-for-Academic-Paper-for-Trading-Strategies-w)
71. [informatica.si](https://www.informatica.si/index.php/informatica/article/view/12899/6712)
72. [pm-research.com](https://www.pm-research.com/content/iijpormgmt/52/2/local/complete-issue.pdf)
73. [projekter.aau.dk](https://projekter.aau.dk/projekter/files/784378028/Masters_Thesis.pdf)
74. [gecf.org](https://www.gecf.org/Portals/0/xBlog/uploads/2026/4/13/2025EditionoftheGGO2055.pdf)
75. [pictureperfectportfolios.com](https://pictureperfectportfolios.com/how-to-invest-like-steve-clark-successful-hedge-fund-manager/)
76. [medium.com](https://medium.com/@navnoorbawa/the-covariance-matrix-killed-ltcm-its-also-how-aqr-and-two-sigma-generate-alpha-d5ef8fae9945)
77. [toptradersunplugged.com](https://www.toptradersunplugged.com/podcast/profiting-from-different-investment-regimes-ft-rob-carver/)
78. [pm-research.com](https://www.pm-research.com/content/iijpormgmt/45/3/local/complete-issue.pdf)
79. [funanc1al.com](https://funanc1al.com/blogs/follow-the-pundits/best-hedge-funds-2026)
80. [dspace.cuni.cz](https://dspace.cuni.cz/bitstream/handle/20.500.11956/200417/120507159.pdf?sequence=1&isAllowed=y)
81. [mdpi.com](https://www.mdpi.com/2079-9292/15/6/1334)
82. [researchgate.net](https://www.researchgate.net/publication/265366427_Hidden_Markov_Models_in_Finance)
83. [pm-research.com](https://www.pm-research.com/content/iijjfds/6/4/local/complete-issue.pdf)
84. [wne.uw.edu.pl](https://www.wne.uw.edu.pl/download_file/6095/0)
85. [medium.com](https://chierhu.medium.com/taxonomy-of-hedge-fund-and-asset-management-schools-of-thought-3f78b7354c04)
86. [pm-research.com](https://www.pm-research.com/content/iijpormgmt/52/2/local/complete-issue.pdf)
87. [pm-research.com](https://www.pm-research.com/content/iijpormgmt/46/6/local/complete-issue.pdf)
88. [theideafarm.com](https://theideafarm.com/research/)
89. [scribd.com](https://www.scribd.com/document/700909978/systematic-strategies)
90. [quantifiedstrategies.com](https://www.quantifiedstrategies.com/volatility-atr-bands-strategy/)
91. [medium.com](https://pyquantlab.medium.com/dynamic-trailing-stops-using-atr-2d3c4e95ddc0)
92. [medium.com](https://pyquantlab.medium.com/an-algorithmic-exploration-of-a-trend-following-strategy-with-regime-filter-and-dynamic-stops-70a7c6d41134)
93. [chartscout.io](https://chartscout.io/stop-loss-crypto-trading)
94. [researchgate.net](https://www.researchgate.net/publication/400742265_Systematic_Trend-Following_with_Adaptive_Portfolio_Construction_Enhancing_Risk-Adjusted_Alpha_in_Cryptocurrency_Markets)
95. [arxiv.org](https://arxiv.org/html/2602.03874v2)
96. [farzulla.org](https://farzulla.org/papers/Farzulla_2025_ASRI.pdf)
97. [repositum.tuwien.at](https://repositum.tuwien.at/bitstream/20.500.12708/210742/1/Akobian%20Liana%20-%202025%20-%20Modeling%20Dynamics%20on%20a%20Canonical%20Neural%20Manifold%20in%20C...pdf)
98. [researchgate.net](https://www.researchgate.net/publication/229445330_High-frequency_Markov_switching_models_in_the_foreign_exchange_market)
99. [repository.tilburguniversity.edu](https://repository.tilburguniversity.edu/bitstreams/2bf5a880-b5dc-4b27-8b4c-368a2b0d0b6d/download)
100. [quantifiedstrategies.com](https://www.quantifiedstrategies.com/hidden-markov-model-market-regimes-how-hmm-detects-market-regimes-in-trading-strategies/)
101. [luxalgo.com](https://www.luxalgo.com/library/indicator/hidden-markov-model-market-regimes/)
102. [blog.quantinsti.com](https://blog.quantinsti.com/regime-adaptive-trading-python/)
103. [wire.insiderfinance.io](https://wire.insiderfinance.io/hidden-markov-models-for-market-regime-detection-d528cfdc3af7)
104. [medium.com](https://kjtradingsystems.medium.com/bull-and-bear-regime-trading-how-to-algo-trade-with-trends-and-not-get-run-over-by-them-b1c39bf2dec)

**Sources:**
1. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHtmfzaJP-SGDFYWiSKGjb0qm_MDPpztU3gleVBXuNSd3TV2TyZZlk5OF76OvRQRRO9DRihlgwuL_zh2cO_sblJS8ztn7PQtNjCjNCt_i_QfkDsIRVWIii4qIt8WedUqCEhiDRQp9GXd14fnSZd-veBz1Q24kVawZaXzL6jKp6Orz43Ktjn8yVICeT-85Q=)
2. [cube.exchange](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQE0fP4jSBc3-iZcBZ0wnO_ZAl0VEQ3d38cGMZwswre9UW9O0M8ukN3pmZTDC_v9tQhfHBEtZ5f_GOkB8_yJE4ZvYmsLMTK3ih2JcCRTrWRq1_-mBDkShwPKCK1uQEZJVUY7sdNTeQarEDkA2xCqNscHxTWJ1ze4CPSVz2GYaCQg8kM_VJCjuPlT6g==)
3. [regimeforecast.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGPUVu0mztWQJINjakdNy8RPFo4l-KU_lTZQLQDc2ezzWR9iWO2stpMkvAgrJ6pQHrAf47GXc3FVLG0Ngtodk7M54XXJaDvgk-wn7tWSz_Nd8W8qhWYL49KWTZJ85Ur5MS49A0QfEkG2ox1faxnBUnsWz1to-zSQ42aZ00=)
4. [quantifiedstrategies.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQG5J14bpCoTzQ3gli9PZAGB9y73WbjAVNnaf1U0C16Fz29YgzWo3s6G09_pelP2B0pgHs7tmy02tqj7Vvwh6UNtp0VpTQqHmdq-0l9TZ7IS3hBse3wUDaxmY0KfZVgpD07Z4PxfOKl0PWExEWJH4GZ_gEOmIwXFMFttqOsP6YAAUNXNOjZLpn0n1MUgwNuEtPQxaJUDAudoXCN1cn_TRyvnE_cRl2mH1ubp_JxiqkQfGsUUXw==)
5. [medium.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGDhtsW3sIjV_-M0GjhPzZ-trpMsSRpp4pn3E-5JBTf4VG_HFov1EL2TRPXUO72S6jmNoFjRNFXK4QvbE4pO4SlyqGTfHMi-AbwhEEiKTgwnVOmbH7mRqB1iZbRFa8nWn4iwQ2qSSXPudcFHgoV-ZSDZdUxj3RyOY1qrgZxIVb4lqiEWUrm6ZlylX4q9WYqqWaMWsuGBvXzzgAOWSytfJiBMUR9caVVaugSfvOZRQJ-hpjt)
6. [insiderfinance.io](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGLzl8ifzRPorsmixzwzJQo1QlEnYUmf8q1lntCR8uEcyk0u2JDc0cURJ5DmqB6tHWo_eFSlV3IYZNh9_q3x_9tovJ17oVWRZ9Xnie71K_fdIoJidIxe7MEM5r3hlnLJTRms5o98taCa0au_NxMWzqsOhK03ZszIxmRG62I5nWC17vUrk6EFxFS25MPnt16k6lylg==)
7. [cam.ac.uk](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGNS5I6RvtZ36S0xjJVdKZ9Ni3IZ_OSu8kq8p-qn9l6mrqsSKJYdIcIT5tz13eEj94o-bXBVBCisE_U6itW3gXa9fLEl2eEHwK3yNCrxeY_O8i7wq7mmjySi6ypuPj3XKaP)
8. [reddit.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEsDi5kOiOALn66Fb4ef0wFIvK0adNzYf86UaxUFkfdyX6YDBBeXPNkWTZU0OBhJ4eo4AVFxRrDkmlYMIxQCR1Go1ml7xZGD6BjQABqgqqfT2L0JeYw38L8ke5_QTjNIa9iIVoQNirqJ-pgyJhPi9hzbDeMXACe)
9. [Link](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGK4A5yu6p6D4HTx1X09hBNjslcikyibu0hIYhiqFZisoUSGGLD2AoXaAZ9d3XT-TVvQfYwRAyQavUiu5id9NqDT-dDwoYCyNZVXk0odLXuZOF0bkNCCTfmm0NMeSFhrfOVf5MFA7ZEXL18XaEu03svz0pqZU1qIEQg9wzvkoOzaREE2ApMwvJufopaDBT2UgMWnFdJc3jk_J8zZI40k1eXZqmmX5qK)
10. [mdpi.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGVPz95hZia7Qrw4wFZ4JbXk1xumW_Jue0M2zOMKIQ2DDh250TgPvqFVpSDJgwU54Je7JY4UIuSp3c34l1NWYP3Dp028A2spI_VUfVzXgLYJa2muwWdmUn5CUmqC9kZ)
11. [mdpi.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGW6ZABoeEDOppZLbtE8empOwf0v0l3T_vt7KRsO6gPCLSJTE7FNAjVlbiY1IxMV01b7syhsfOh83LohdCu6k3oBeVOuvlpI01NesuBvPKVLj72scZzWTSH0nemXm8qZQ==)
12. [semanticscholar.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEXqKNKjXp1DlChtHqhJd5I5GNBC6x2tXolKG2se9-Do5KVKHxYjeXCYlIi7JHmpfygiXUYh4SH_lQi-RLd1E71mfh36irD2jKCiVpW1COxF1HBGnKAbQD87DPprOp94lO2hXHwTSv6wbRCethQ20lTj5B1HBUHIvAOmFafcX3xvgudCQs=)
13. [dtu.dk](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGKQX53hcY7vfgDKj0M73bPbz-gCF_jt1sTVR_fTVfzLTPnNrZDBgSjt6hlNmAxDtsR3DxeYajmoRs-PeLevFFGGg7Ep3STEXHT9-ypqfPt2UyaWZslYdTJWGUVvwOcCwjMTgSFZ6YwDcJ-De9RCSKAEDeFjufAyh6Oig==)
14. [pm-research.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGbj3caP84gSJuygcU1ZAyvw0IAFTUp_0lA5hBxSrsZo39vh-Os541GpHP1hAowEUT2jWUIb0N4klW5KCdZkZ0YvpL8O_jljUrHP-oOZW73aCVeQ6vWu954j2LM2Q8coN7PwOjWHvgftw3IUeQihJki_5le3cof_9pXsnzdS3c=)
15. [medium.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHz7qV2xWLoYtsqtux9rLKMmxjvrt6R7j9iLvF8-Ijk8jDvlVzJLsXoK4imNyaTts3ET-SuU37SweNCpDGw1r73VjmJ-yy0QeG03HPCQx1ze4Oc-3sE0HrMGakAZSd4Y2FQLVzw80Wg9IyYNzQLR6WeZ5cFWNtRKU5pPgG0NBu-CNscsN0nAdAgLpjVtm50FFqgRpZVKw-kV8ktWaO9NRMvzvHPbB1pFg==)
16. [mdpi.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFTeKsyBGwqTEkyv4-5i_LNohXeklEkTz_NaaR_2AXrSePum_oZM5O7kARv-JISJaqtpUamsLTCY_3-mOSkc7-VqvDmYdyqZ_va2ezWzc5PiCqQ_hnjehfaFGlS7wW8)
17. [cuni.cz](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQF8KjC0L_5BPGVGr-K8ePoNW9ZxprcB78sZmClR1htAMigWOB0sEhhPXRivY5pLqD0mkxX0YQFUh4MXLGUC5h1RVN7LPNFgnM9T-zVaYxsAGusrhTXVgBL47DGWtkQ0QEoof7Hl0StHIS3CWhRlrZtQZTwlJVgxwNcobRvbl1Q_YAAIjIu-Y0mPinwCQqf3nVgrhcOOGsA=)
18. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGr0otJJRHbrhifl1AccYQ-cZ4gdq3dalLGr3DN1prJlEVYoLKc0DerlJuYhU3hVv4btpAzjMD2EDUoJ-2bjc8VUS-305igiFO9yI_pRImMOdRvOSEN2i-KVg==)
19. [farzulla.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQH4O9PUoz5PBTiAulW6vTyjKY-7_FVOM3BTh2StTr6VDP2ThlYg06JoJX2PfoCVOOTfskGdDaJkMqZvJxUxe3UKi2HNC4-cbwS8FBH7rw9M4p8uT-EdMAWS7Dr1Hfkrzusa6U1t_6leHw==)
20. [wordpress.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHSLmSe1EAQDNobqw90MydS9BaY33CDFIbA8ZeleAq6rchGQt8rlx_9ZRnH94IEU1ZkxKRR1wn2OD9WjHwqloT_RJe2NNMyOY5Fubwr_ssNZ90PVl9Wjrq8l2LyjdqXoi2JgXTSNHp8gV4m1acx3z7BVmrfBtL_1Z_v1PVDkCIlifPKYBspW8ah3NH-BN8oidNx7wL3aPzpMoIdhXIQL323YlLFJk9zcWWj_3ZbGopbFhtL)
21. [Link](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGa8gdnpZFiew3zZ8aeJUnPnhBFFuUf6l922nLfTR0fAFeWxf07BXuWKn8dLHH5BlDtCbSVGAjEk2zyM6fg3W0q75vPA9ReleAF6dRNxNLpUQosIy5J96Kxlt0J5g==)
22. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHYdLiBImTwcqIsPqG67-PJVYLV4hLiUEVRSMO3nH3JPGB4WaFmlMS5Nik-nPMAwmyknT0a8pz57OcH_X-Pwo_vCrAwiIH5dFd2lrhhxjzOl0qw249I0BwibQ==)
23. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEXXZ0bdmUHCPm2TTBO_wmhwvp0oKBp4071org-ssxWVHlRol-CoXlNRrTYcdjN4ocWq6g6onR5NCvAT5hxQkXqDFWTaTvBZztxMI1JZHJ0sKE39CLhtg==)
24. [repec.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGWf5VPiwPTk-i3KjtUGfy-XQ3asxiMryue9bW292PvoXZwkLSZCSaxlSSVe20hWo1VpNvJXlZ8oikufOFPk4vEZ5J8-MYuvoHNueRUtA_gl8LTtLDJSagV3B2ZeBYV_N7hkw5TVRXQX21L)
25. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHhmAiuN5HxxV9gonCy_s5DbHverLAFwuRUuElOdrFpQw8YZJ8i_ve4XxUg1zDb7vp8a5s92MoSOCxziud-twA6oNfUUM568MjKOQB5V3QzwJA-HpgudA==)
26. [Link](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGN_nphUW8mIeXmNBe6cqy4J1dWqqqMzAUJ6N29aQfndDxTzYm0oPmFuPVfTr3TSKEcTWc-u3K-6EFScNT13yFdKtcWOkgQtVOerLKdAurIgjLcrJhbPuJzFQ==)
27. [theideafarm.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGXdEUDhumeR8-ua37VihQzArDlis3eBtbrhd6xeBRqdIc-akXSGSnYrX5XSUaM7raRby_WkdNnOniU-mUDms63o77bJRHxjRkNhpbRE0ZCjLfwPtA1dMmT_L1535Onf2B2YPy_VFz5J299h94=)
28. [toptradersunplugged.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEkxgAVoQl1tXK05IBAvlz6_0XyneCr8yXJYyOdlowKRbESFUgfduX1B2DzwLRLzQ822KQDIkg_XIPTPMmd9smZspJIY7qOvXDMUFZAhNUGZm6ODgjkp5omk9S0gwGGWP5t_kc8olXW7Eudb09xvrIK5LKWu-Z_ytDGcLJ9HwFKwryiQ2Hh9FPNtiRgmrJhva7QkKfefYAJ7YDgcH0=)
29. [man.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFocmt-w_uriXUuPdWNbJYZ5jPqM_uKgfq3tpOT-6BRuM-c_XvOoh5nod9duSFVSgriGB7-EJcxyhdlcui91DAe1saAWgUDSOUTDKge_Ll4_5j8EE8ScIH432ewBemILq0TC6XSqBTft1tLb6rE8sLNy161WViR-GAUWJ8nJTYaZd9B4nZ5l8qRWu5aUYc=)
30. [pictureperfectportfolios.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGpL_Nl8AzhvFQArFlJ4eWiQynp5zLI-SCM8imhwTQ1-l9O3lh8O9eokbqdXCXT38Abg-k12wdS5cIzE-Hw2Ump1V2_pzssZxDh-dhB6UqFlBc2zqycq6IIlPJLvw5NXUfO_iT6nZvebicKIxY_M7aF_srXH1EIB9sGD1M1MQo8idY1HpTfMMBzgTwl12pxKvXe3n_ywlQIcQ==)
31. [turtletrader.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQG06ueHqzA0a52bqageB87sziVRnHd-xR8e7VnQmHkrweh1yZHbOvAZcjFDbssuB3wm4GdJ5QJcWGJdOaQlNGolpmnef44aXG2jJ-ZyzIlXOlKYlkjpETDGa6ZQtoWLx1LD4ZUdf_ajLaT52WI=)
32. [man.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGZXacBGsWgnA-kBzB-3y2hx8MJHWFqmOuqVxJQMEVi6kJ7_Qm2qOQzX98YWpkvz0HyhiZWZvUglzVzd_Xr_UqfHO18Pc0MZ2hS25Vqz33ruumMVebdsbLAUCUDNxE_0z0uGf9ldtEqhP6LvdyXiJEbdUpBSc9VpnfraMVAfLPZcQ==)
33. [quantifiedstrategies.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQG_KI10J3NWRC6OwygRjHwgPuezhCIxi0uHPFXlnfbQ2_vu0O9bLZTuaSvIWgRUXyAzPg9pOTam0FJpwjVBCRxps0jNJrmKNZPtUrxzNDwujXI8W2BGG2R2HT97VYXIswBUwZeJrTP5jfUr_6_SdxHLaDj4IbIDaYbg)
34. [chartscout.io](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFlU_SRihbUhOS9WW7VBHKYAlIS6Jip83BJCnmF4VPcqa4C98gEz0D2TDOrKtmA-hjN3aBCEWudJljOfoFXPFkhhtrgc_bmkilBL2s5f-k5ASpsxeYQpvcNULDwwmfwvED9jmsh)
35. [luxalgo.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHL6Jhrfmnr838crxXKiW5C3Lob9gTIuE4FCcM2Xw5TRFvngbSTxr_Z09QYyvhWwmbu0ftQJhK3sHYAr6K8y8h-HWdaxzfG337rHf_mgkH-a-Rfz_g-Ai00_SyAPja1IqxtcGjAdiP8DbxT2gld3qU-IXmT8elFRBLwjaRZcb6PlJt2yg==)
36. [medium.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEOeNxA1tkQIPdp-3nAorvmvrPfCyj1ugHYq5WyizHAycj5onlBCHAA1Bk72XzVm_5dCeSguD-O0ChR__AK620H9KSmDRjWr27Uq2mMIo8YfpfPHN9Q5Z_M1q8uTZvDpbyZEVXwQYemj5dp_81uhJqpntgpzFq-iErWmIpqqWsfomE=)
37. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQH_6vRlKJEKVxuZNzR3miC6n-YjYNQA-o08MD-Oq1fo6uVmMcec415NWhL84WKn0HFZP-9phv4m0ejA1s0vB5n2_uoZKGDZHcs5ElswHTGl_urTZvvutbqEmBtpGGek-BlSgMma78vDh1kQYseMqOd-XvRBysQksKCp2skYjmoynt2SyBx-OO7Jbph7uqJ2xzimstYyIQIj2TM5k8n4tz0gKCnYu9Q9sEZwVCgIDpaotZw-FeMsuy5XbIgrUP6DaUNVzOHgw-hw4kLlhKUZFV7LKiYL8xk8tjcPs8VSt7GRFjE=)
38. [medium.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFAJfXIPKd-I0a_Omn0f8Ur33p7H26FP7jL7YINXyeW7oJ2O41-Pfnth07lbzTTxBle4FGWG1AXrWCIKSOqa448VCFxdepVTJLSwUEGP_V63YixjKSdZdRnoacee35TjMB70zyAuPahJVZbQXgcBc_KduxUrCHA-i0AbPRFac9_ez0OgNqRbLBV37lPKl9aeAniUfypgPfQY7YIwTGL4p2WVpfbQgk31GMZc8J3FypLyudceZrXiAItjweStHL6oA==)
39. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQE0I4KqQwO8UuZxuELdcRrmXWNAf5ubypjZ5spnBfKhN2o61ryM1nrrIl6oKCEPpftigtwOL-3uSDG2pWxvwu2cduuC7C6Pms6rEow6GrDuP3-8wjY_RrWR_9z63DjnTTx-Oj8YiMMgO35b7GD-s4M2hQftzns2aTeqWoaTS4EZYXD1YAn23Iaq_Fbect9pLQkzGvwyJc3o6G4QHkjuHrQ_Vv_bqw==)
40. [polimi.it](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFPEySe97lSkP7rv9fvw9PoVKJr0gg3juR4xMlj2rSCr02ToWAcMNkRZdfX20m5QiQmVybsg1rpWN7_zD8yHOCGiUAr3fbKpgKogUvM-M8LtRb-G-9Xlv2BKzIDHqCNck-0Ik0iLbLrg1zZ1wmT00F2CnXMn1DCgrxHDWmpfX0sMVK3xjqD_rFJAArjzq0TAM2m9rxz9_iCYai59nJV_Fqs1MGeAZ0ksg==)
41. [medium.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHvU1hCGsRov75qJzam_MnxN6bxrdsmgA__wQvYxtHkS5s6DcQHq3UOXOPQh8URLVQSwgUBWgjt7RbhmWZbu8xMzIIvHwe9AOiD0kwVPlHT6FpR_5-qnjZPcvFjRvNxrn6LhUVmtP2Se_AhTTbeaE1WYwUSzhzKrv4oalhDVhZbGlkgJrJuXD1iHabwiwcIN-mIKxrmuf9DlUzSaPp1AI9K0KWb_w0u6GFebH3DK3eAsdNzDBsC3rLIUWK7MHHK)
42. [hedgenordic.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGtX-uVAvg3XLrTixDxs51HQD1vGaVBd_VSl4KZn124Ou57f2-I7NpitGbmCqYwu6rQ6n8X1gJAJpu4hCL5gs1u-gMjKVh5OlyrjyYyVpCtWrxf2wxZ3MVjKlR7TF2_-VEkwGNe5UDDqNBm5bKouK8vjJ5wTr0XrvjT0I6xyVdyqC9F)
43. [mdpi.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQF-i-4qr4fvc3u6861y_jzKLEDP5MfywAXBo4bzpW53hcmrhqBsmgsMp33Kkhhbun_YU1QWu-wvhIcQkv4Js2OfNt1LacghWFVq1Q5iTXoO_2y6tRAI5ESmI1TvhOc9FQ==)
44. [aimspress.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEIvCCGs6dWFGzkXKrnSyWvp-kyC7gS_UvBufj6HM6WoTUMqfM4eNp6a6uzquciDKUDjKM5ox8hF0tgu0WXx7jwMnH9q-cXw7UG9O2jsaIpkk4rABAMGEaEySWYfhGaURFxfle6Kf2xQjfvxvjeprbC4dACfr-N2GupHhKXndk=)
45. [pm-research.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEIT4mpk7RxC-66YyF_XFWYILjT4gwzwHruKLvv5EW0ZYdFfnCjnAvXKY92PgGTQxpBi0Rvo--AmGtU_t5iLlOcnpf8g3FgKXjSgSycbYsU6_v7yDRxT9XCYUOt3Q91UzHpk2R0sXVIXpeMLwJOPL50lRDBRbCQcQYMsEBYtGy6RmWj)