# Empirical evidence on RSI and MACD as momentum filters

The pursuit of sustainable, risk-adjusted alpha in global financial markets has long relied on the quantitative measurement of price momentum. Originating from the foundational tenets of early technical analysis, momentum oscillators such as the Relative Strength Index (RSI) and the Moving Average Convergence Divergence (MACD) have fundamentally transitioned from manual charting heuristics to integral, systematic components of complex algorithmic trading systems. In the context of modern market regimes—which are increasingly characterized by post-2020 high-frequency execution algorithms, zero-commission retail trading environments, and the unprecedented structural volatility inherent in cryptocurrency ecosystems—the continuing efficacy and statistical validity of these standalone indicators require rigorous empirical re-evaluation. 

Before proceeding into the substantive analysis, a structural note regarding the research scope is required. The requested companion files, identified by the local designations "e685" and "e686," appear to be localized user documents or proprietary internal references that remain inaccessible via open web search repositories or academic databases [cite: 1, 2, 3]. Consequently, to fulfill the analytical mandate without compromising the integrity of the research, this report explicitly pivots from these localized documents to establish a standard, rigorous methodological baseline for momentum indicators. This baseline is constructed utilizing a broad corpus of recent (2023–2026) peer-reviewed finance journals, SSRN preprints, and institutional quantitative whitepapers from leading entities such as AQR Capital Management and Robeco [cite: 4, 5, 6]. The ensuing report delivers an exhaustive synthesis of contemporary literature, evaluating the intersection of traditional momentum filters with modern machine learning paradigms, addressing systemic biases in quantitative backtesting, and examining the theoretical friction between technical momentum and the Efficient Market Hypothesis.

## Methodological Baseline and Theoretical Mechanics of Momentum Indicators

To critically evaluate the contemporary utility of technical momentum, it is essential to first establish the strict mathematical and theoretical mechanisms governing the RSI and MACD. The momentum anomaly, which has been extensively documented in institutional quantitative finance, generally manifests in two distinct forms: cross-sectional momentum, which evaluates the relative performance of assets within a defined universe to bet on relative winners versus losers, and time-series momentum, which evaluates the absolute historical trend of a single asset over time [cite: 7]. Technical indicators predominantly function as filters for time-series momentum, seeking to mathematically quantify the velocity, magnitude, and persistence of directional price movements.

### The Relative Strength Index (RSI)
Developed by mechanical engineer J. Welles Wilder Jr. and introduced in 1978, the Relative Strength Index is a bounded momentum oscillator that measures the speed and change of price movements [cite: 8, 9, 10]. The mathematical core of the RSI involves computing two exponentially smoothed averages over a specified lookback window, typically fourteen periods: the average of upward price changes (Average Gain) and the average of downward price changes (Average Loss) [cite: 8, 11]. The indicator calculates the relative strength (RS) as the ratio of the Average Gain to the Average Loss, which is then normalized on a scale from 0 to 100 [cite: 9]. 

Traditionally, RSI readings above 70 denote overbought market conditions, indicating a statistically excessive positive momentum that may not be sustainable, thus signaling a potential negative mean-reversion or correction [cite: 8, 9]. Conversely, readings below 30 denote oversold conditions, indicating potential exhaustion of selling pressure and an impending positive reversal [cite: 10]. However, modern quantitative application of the RSI extends far beyond these static, absolute thresholds. Advanced methodologies rely heavily on divergence analysis, where the failure of the oscillator to confirm a new extreme high or low in the underlying asset price serves as a leading indicator of momentum exhaustion [cite: 10, 12]. For instance, a bearish divergence occurs when the asset price achieves a higher high, but the RSI prints a lower high, suggesting an underlying deterioration in buying pressure before it becomes evident in the raw price action [cite: 13].

### Moving Average Convergence Divergence (MACD)
Conversely, the Moving Average Convergence Divergence, developed by Gerald Appel, operates as an unbounded, trend-following momentum indicator [cite: 8, 14]. While the RSI normalizes data to identify statistical extremes, the MACD mathematically expresses the evolving relationship between two exponential moving averages (EMAs) of an asset's closing price. The MACD line is conventionally calculated by subtracting a longer-term 26-period EMA from a shorter-term 12-period EMA [cite: 8, 14, 15]. A 9-period EMA of the resulting MACD line is simultaneously plotted and functions as the "signal line" [cite: 14, 16].

The MACD generates actionable trading intelligence through three primary mechanistic interactions. The first is the signal line crossover, which indicates short-term directional shifts; a bullish signal is generated when the MACD line crosses above the signal line, while a bearish signal occurs when it drops below [cite: 14, 17]. The second mechanism is the centerline (or zero-line) crossover, which serves as a macro trend filter; when the MACD moves from negative to positive territory, it confirms a broader structural transition from a bearish to a bullish regime [cite: 17, 18]. Finally, the MACD histogram, which visually represents the mathematical distance between the MACD line and the signal line, measures the acceleration or deceleration of the underlying trend. Expanding histogram bars suggest strengthening momentum, while contracting bars provide early warnings of trend exhaustion [cite: 14, 17, 19].

While these mathematical mechanisms provide a highly structured and objective approach to quantifying market psychology and price velocity, their seamless transition into systematic, automated trading algorithms necessitates strict adherence to robust quantitative methodologies. Without such rigor, these indicators are highly susceptible to generating statistical illusions that fail in live market execution.

## Systemic Vulnerabilities: Data Snooping, Overfitting, and Survivorship Bias in Backtesting

The primary vulnerability of technical momentum strategies lies not necessarily in their theoretical foundation, but in the empirical validation process utilized by contemporary market participants. A persistent and severe critique from institutional quantitative analysts is that the historical profitability of moving average crossovers and momentum boundary rules is frequently an artifact of deeply flawed backtesting protocols [cite: 8, 20, 21]. As computational power has democratized and proliferated, the ability of both retail and institutional traders to rapidly test thousands of parameter combinations has exacerbated the multiple testing problem, leading to severe data snooping and overfitting biases.

Data snooping bias occurs when researchers or algorithmic developers unconsciously optimize indicator parameters—such as iteratively adjusting the RSI lookback period from 14 to 9, or tweaking MACD EMA lengths to non-standard variables—so that they perfectly align with historical market anomalies [cite: 20, 21]. Because these specific historical conditions are statistically unlikely to repeat with the exact same frequency or magnitude, the resulting models perform exceptionally well during in-sample testing but fail catastrophically when applied to live, out-of-sample future market data [cite: 20, 21]. Recent quantitative literature emphasizes that financial data is inherently noisy, nonstationary, and structurally distinct from the clean datasets utilized in standard machine learning classification tasks [cite: 21]. Standard hold-out validation techniques are often insufficient to guard against this multiple testing bias, as evaluating strategies on overlapping historical data artificially inflates false positive rates and provides a mirage of predictive power [cite: 21]. 

To counteract this pervasive issue, modern quantitative frameworks, such as the FINSABER backtest architecture discussed in 2025 literature, employ rigorous rolling-window evaluations [cite: 21]. By dynamically shifting the evaluation window and step size across diverse and constantly changing asset selections and extended time horizons, these frameworks ensure that performance metrics are robust across shifting market regimes, including varying interest rate cycles and volatility conditions [cite: 20, 21].

Furthermore, researchers evaluating RSI and MACD must aggressively combat survivorship bias. This is a pervasive methodological flaw where backtests are conducted exclusively on currently active, successful assets, retroactively ignoring companies that have gone bankrupt, been delisted, or acquired under distress [cite: 20, 21]. Because delisted assets generally exhibit severe negative momentum and technical breakdowns prior to their failure, excluding them from the historical dataset systematically overstates the historical returns of trend-following and momentum-based strategies, while drastically understating the associated maximum drawdown risks [cite: 21]. 

The structural transition to post-2020 zero-commission trading environments has also fundamentally altered execution assumptions in strategy backtesting. Retail algorithmic backtests frequently assume frictionless execution at ideal closing prices [cite: 20]. However, institutional quantitative whitepapers, including extensive research by Robeco, consistently note that capturing short-term trends via factor momentum promises theoretical alpha that is meaningfully reduced, or entirely eliminated, when realistic transaction costs, bid-ask spreads, and slippage are accurately factored into the empirical models [cite: 4, 20]. In highly volatile environments, the rapid generation of RSI and MACD signals can lead to excessive portfolio turnover. Consequently, modern evaluations of these indicators must rigorously adjust for these microstructural frictions to separate genuine, exploitable predictive power from theoretical optimization.

## The Efficient Market Hypothesis Versus Behavioral Technical Momentum

The theoretical justification for utilizing historical price-based momentum indicators remains a subject of intense and ongoing academic debate, primarily structured around the classic Efficient Market Hypothesis (EMH). Initially formulated by Eugene Fama in 1970, the strict interpretation of the EMH, often associated with Random Walk Theory, posits that asset prices instantly and fully reflect all publicly available market information [cite: 22, 23, 24]. Under the weak form of this framework, historical price data and trading volume possess absolutely no predictive value regarding future price direction, rendering the entire discipline of technical analysis mathematically futile [cite: 24, 25]. Institutional critiques aligned with the EMH argue that technical indicators inherently ignore macroeconomic fundamentals, fail to account for corporate intrinsic value, and demonstrate a dangerous over-sensitivity to random market noise [cite: 24, 26].

Despite this formidable theoretical opposition, empirical evidence gathered over decades consistently demonstrates that time-series momentum delivers positive, statistically significant returns that are relatively uncorrelated with traditional equity and bond risk premiums. This suggests that momentum acts as a viable risk factor or behavioral anomaly that persists across global markets [cite: 7, 27]. To reconcile this empirical contradiction with efficient market theory, contemporary financial literature frequently references the Adaptive Markets Hypothesis (AMH). The AMH, advanced by Andrew Lo, reframes financial markets through an evolutionary and biological lens. It acknowledges that market efficiency is not an absolute, static state but rather a fluid condition that fluctuates based on the behavioral adaptations, competitive dynamics, and natural selection of market participants [cite: 24, 28, 29]. 

Behavioral finance provides the transmission mechanism that allows momentum to persist. Within this framework, the recurring patterns identified by the MACD and RSI are theorized to be mathematical, quantifiable representations of persistent human cognitive biases—specifically, delayed overreaction, anchoring, and herding behaviors [cite: 24, 30]. When new fundamental information enters the market, participants do not instantly price it with perfect, rational efficiency. Instead, initial underreactions and gradual institutional accumulation create slow, directional trends, which are efficiently captured by moving average derivatives like the MACD [cite: 30, 31]. As the trend becomes obvious, speculative retail herding drives prices to irrational extremes, which are subsequently captured by the RSI overbought and oversold boundaries, signaling impending mean-reverting corrections [cite: 30]. 

However, recent high-frequency studies published in 2024 reveal that while these inefficiencies exist, the weak form of the EMH reasserts itself rapidly in modern regimes. During periods of extreme geopolitical or macroeconomic volatility, the predictive capacity of rigid technical rules frequently breaks down [cite: 11, 32]. This dynamic suggests that while behavioral anomalies can be exploited via RSI and MACD, the inefficiencies are increasingly short-lived and highly regime-dependent, requiring sophisticated timing mechanisms rather than passive execution.

## The Institutional Perspective: Factor Timing and Quantamental Integration

Insights from leading institutional quantitative asset managers, including Robeco and AQR Capital Management, provide a critical macro perspective on the modern deployment of momentum filters. Institutional quantitative models rarely deploy standard 14-period RSIs or 12,26,9 MACDs in pure isolation [cite: 4, 5]. Instead, institutions view these tools as rudimentary, specific expressions of the broader "momentum factor," which is systematically integrated alongside other established fundamental factors such as value, quality, low volatility, and size [cite: 4, 5].

Robeco's extensive research into factor allocation emphasizes that while factor momentum—the practice of identifying which specific style factors are currently trending—is theoretically predictable, efficiently harvesting that predictability in live markets is incredibly difficult [cite: 4]. A pure, unconstrained momentum strategy necessitates constantly rebalancing the portfolio to chase short-term winners and dump losers, resulting in severe portfolio turnover. The associated transaction costs, market impact, and tax drag heavily erode the theoretical alpha [cite: 4, 27]. Consequently, institutions often rely on robust baseline approaches, such as risk-parity or equally weighted (1/N) allocations, to capture long-term structural premiums with significantly lower turnover, relying on momentum strictly as a secondary confirmation filter rather than a primary allocation driver [cite: 4].

Furthermore, AQR and related quantitative literature advocate for the integration of momentum metrics within more advanced predictive frameworks, such as online Classification and Regression Tree (CART) models and Support Vector Machines (SVM). These models train on efficient frontier coefficients and latent volatility regimes, utilizing momentum indicators alongside economic data to dynamically shift asset allocations rather than executing binary buy/sell rules based on static oscillator crossovers [cite: 6, 33].

The evolution of emerging market (EM) strategies at Robeco underscores this shift. Over a 30-year operational history in emerging economies, the firm noted that momentum historically performed exceptionally well when markets were heavily dominated by retail investors prone to speculative "herd mentality" [cite: 34]. However, as emerging markets have become increasingly institutionalized, the frequency of these irrational, retail-driven price extremes—the exact scenarios where RSI boundary strategies excel—has diminished [cite: 34]. To adapt, institutions have embraced "quantamental" investing. This approach merges quantitative momentum and volatility signals generated by computational models with deep, fundamental human analysis of corporate earnings, policy documents, and macroeconomic environments [cite: 6, 35]. Empirical tracking by Robeco demonstrates that combining systematic quantitative models with fundamental stock picking significantly elevates the Information Ratio (IR) of developed equity portfolios (achieving IR improvements up to 158% relative to average fundamental baselines), successfully mitigating severe drawdowns and capturing the optimal intersection of intrinsic corporate value and positive price velocity [cite: 35].

## The Impact of Machine Learning and AI: Are Standalone Indicators Obsolete?

The explosive proliferation of advanced artificial intelligence models, including Long Short-Term Memory (LSTM) networks, Convolutional Neural Networks (CNN), and Large Language Models (LLM), has fundamentally disrupted traditional quantitative technical analysis. A critical question dominating contemporary research from 2023 to 2026 is whether these sophisticated models render standalone heuristic rules like RSI and MACD completely obsolete, or whether the indicators maintain utility as engineered, structured features within deeper, non-linear neural architectures.

Empirical evidence across modern market regimes suggests a highly bifurcated reality dependent upon the time horizon and frequency of the data being analyzed. In highly constrained, ultra-high-frequency trading environments, standalone technical indicators demonstrate significant mathematical degradation. A comprehensive 2024 study analyzing minute-level, high-frequency historical stock data for the SPY (S&P 500 ETF) evaluated Random Forest Regression (RFR) models augmented with traditional MACD and RSI features [cite: 11]. The base RFR model, utilizing solely raw price data, returned a negative yield of -2.40% over the volatile testing period from April to September 2024. However, incorporating the MACD as a predictive feature further degraded the return to -2.90%, while the inclusion of RSI yielded -2.50% [cite: 11]. The RSI model exhibited a deeply negative R-squared value (-0.017) during testing despite high training accuracy, highlighting catastrophic out-of-sample overfitting [cite: 11]. Feature importance analysis revealed that primary raw price features consistently outperformed the derived technical indicators. The study concluded that the inclusion of these traditional metrics did not yield proportional gains in predictive accuracy relative to their added dimensionality, actively exacerbating overfitting challenges in high-frequency algorithmic contexts [cite: 11].

Conversely, in lower-frequency swing-trading, sector allocation, and macro-forecasting environments, machine learning models that treat RSI and MACD as contextual features demonstrate profound and consistent outperformance over buy-and-hold benchmarks. In the cryptocurrency sector, a 2025 study demonstrated that an LSTM neural network integrating EMA, MACD, and Average Directional Index (ADX) data achieved a cumulative return of 65.23% in under a year, vastly outperforming baseline technical strategies and simpler machine learning frameworks like LightGBM [cite: 36]. The non-linear nature of deep learning allows models to capture complex, multi-dimensional interactions between momentum, volatility, and trend indicators that rigid, linear, rule-based systems simply miss.

The integration of Generative AI and financial LLM agents represents the absolute frontier of this evolution. Advanced frameworks, such as IndicatorAgent and dynamic multi-agent systems, utilize large language models to reason over technical data streams simultaneously with unstructured text [cite: 37, 38]. Rather than executing blindly on a mathematically triggered RSI crossover, these agents contextualize the momentum shift against prevailing macroeconomic sentiment, policy documents, and geopolitical news flow [cite: 37]. 

A highly illustrative 2024 empirical evaluation of LLM-assisted trading in the commodities market (specifically gold) revealed the vast gulf between standalone rules and AI reasoning. The study tested multi-agent LLM systems analyzing MACD, RSI, and Bollinger Bands alongside fundamental market data across a 294-day evaluation period [cite: 39]. The multi-agent LLM systems achieved extraordinary total returns ranging from 33.1% to 61.2% depending on the temperature setting, alongside Sharpe ratios exceeding 1.0 (peaking at a "very good" 2.13) [cite: 39]. In stark contrast, the standalone rigid MACD baseline returned only 17.5% (Sharpe 0.82), while the standalone RSI strategy executed a mere single trade, returning 7.12% with statistically unreliable metrics [cite: 39].

[image delta #1, 0 bytes]

 This data definitively demonstrates that while the underlying indicators contain latent predictive value, the dynamic reasoning and contextual filtering capabilities of modern AI are increasingly required to extract that value with superior risk-adjusted precision.



## Geographic Evaluation: Equities in Developed Versus Emerging Markets

The behavioral characteristics of momentum indicators are intrinsically tied to the market regime, underlying liquidity constraints, and participant demographics of the specific asset class or geography being traded. Recent cross-sectional geographic evaluations reveal profound variance in the reliability of RSI and MACD between developed and emerging economies.

### Developed Markets: The Decay of Traditional Rules
In highly mature, developed equity markets, the efficiency of standard technical rules has largely decayed due to overwhelming institutional participation and algorithmic arbitrage. An extensive academic event study conducted on the Swedish stock market evaluated the constituent stocks of the OMX Stockholm 30 (OMXS30) over a ten-year horizon (2014 to 2023) [cite: 25]. Utilizing an OLS market model to establish expected returns over a 150-day estimation window and a 14-day event window, researchers tested eight specific variations of the RSI and MACD [cite: 25]. 

The findings firmly supported the efficiency of the Swedish market against traditional technical exploitation. While certain extreme variations (e.g., RSI bounded at 80/20) generated statistically significant Cumulative Average Abnormal Returns (CAARs) in the immediate post-event window, the returns were highly inconsistent across buy and sell vectors [cite: 25]. Furthermore, the gross alpha generated was deemed fundamentally insufficient to offset standard institutional transaction costs [cite: 25]. The study robustly concluded that conventional, rigid technical trading rules are no longer reliable predictive tools for predicting future price direction in efficiently priced developed markets [cite: 25]. 

### Emerging Markets: Exploiting Behavioral Inefficiencies
Conversely, emerging markets—characterized structurally by higher retail participation, comparatively lower liquidity profiles, and more pronounced behavioral herding—continue to exhibit highly exploitable momentum anomalies. A comprehensive 2025 quantitative study analyzing Indonesian state-owned banks within the LQ45 index compared the absolute accuracy and net profitability of the RSI and MACD over a one-year observation period (August 2023 to July 2024) [cite: 29, 40]. 

The empirical divergence between the two indicators was striking. The RSI demonstrated a remarkable signal accuracy level of 97% (achieving 31 successful signals out of a total of 32) when identifying overbought and oversold mean-reverting reversals [cite: 40]. However, precisely because of this high threshold for precision, the indicator generated very few executable signals, resulting in relatively low absolute cumulative profit [cite: 40]. In stark contrast, the MACD functioned as a high-frequency trend-capture mechanism. It generated 166 signals with a significantly lower accuracy rate of 52% (86 successful signals) [cite: 40]. Despite the high frequency of false signals and whipsaws, the MACD successfully captured the bulk of sustained macro price movements, yielding more than double the absolute profit of the RSI strategy [cite: 40]. Statistical evaluation utilizing the Mann-Whitney test yielded an Exact Sig. (1-tailed) value of 0.522 [cite: 40]. Because this p-value exceeds 0.05, the researchers concluded there was no statistically significant difference in their overall performance validity, but rather a divergence in their functional utility: RSI provides high-probability precision for risk-averse allocation, while MACD serves as an aggressive, high-yielding trend-capture tool [cite: 40].

Further emerging market evidence is found in the Indian IT sector. An evaluation of major tech equities (TCS, Infosys, Wipro, HCL) from 2023 to 2024 assessed the Extended Internal Rate of Return (XIRR) comparing MACD centerline crossovers against signal line crossovers [cite: 41]. The study proved that centerline (zero-line) strategies—which act as macro-trend confirmations—consistently equaled or outperformed the more sensitive signal line strategies, generating robust XIRRs ranging from 20% to 50% across the sampled equities by effectively filtering out premature volatility noise [cite: 41].

In the highly speculative Chinese A-share market, researchers successfully addressed extreme retail volatility by developing a sophisticated "trend-momentum synergy" strategy [cite: 42]. Rather than utilizing static historical thresholds, the quantitative model dynamically adjusted RSI boundaries based on real-time readings of the historical volatility index (HVIX). For example, when volatility spiked above 20%, the RSI overbought/oversold boundaries were dynamically widened from 70/30 to 75/25 [cite: 42]. This dynamic adjustment was strictly combined with a MACD crossover requirement and a multi-day signal buffer to prevent false algorithmic executions during chaotic intraday whipsaws [cite: 42]. Tested on CSI 300 constituents from 2015 to 2025, this synergistic, adaptive approach significantly outperformed both standalone indicators and retail buy-and-hold baselines, maintaining an efficient portfolio turnover below 200% while providing superior risk control across bull, bear, and sideways regimes [cite: 42].

## Asset Class Divergence: Forex, Commodities, and Crypto Markets

Beyond equities, the structural behavior of momentum indicators varies drastically across distinct asset classes. 

### Foreign Exchange (Forex) and Commodities
The foreign exchange market is the most liquid financial market globally, heavily driven by macroeconomic sentiment, interest rate differentials, and geopolitical events. Recent empirical studies underscore the necessity of pairing technical momentum with fundamental and sentiment analysis in Forex. A 2024 study evaluating the GBP/USD exchange rate highlighted that while indicators like RSI and MACD are valuable for confirming trends and generating secondary trading signals, their standalone efficacy is severely degraded during volatile, rapidly shifting geopolitical environments [cite: 43]. Technical momentum in currency markets is highly reactive; thus, researchers increasingly utilize Natural Language Processing (NLP) to gauge forward-looking implied sentiment from news headlines, which frequently captures up to 50% of the variation in returns before technical indicators can react [cite: 22, 32, 44]. Consequently, MACD and RSI are heavily deployed as confirmation filters rather than primary directional instigators in institutional Forex desks. Similar limitations exist in commodities, as previously evidenced by the gold market study where standalone momentum struggled without the contextual reasoning of LLM agents [cite: 39].

### The Cryptocurrency Paradigm
Cryptocurrency markets represent the extreme, chaotic end of the modern financial volatility spectrum, severely testing the mathematical limits of traditional momentum oscillators. Because crypto assets frequently experience parabolic, speculative bubbles driven by retail euphoria, standard 14-period RSI settings frequently remain pinned in "overbought" territory (above 70) for weeks or even months at a time [cite: 13, 45]. Relying on traditional mean-reversion signals in this environment is highly dangerous, particularly for short-sellers who face catastrophic drawdowns during uncorrected trend extensions [cite: 13]. 

To adapt to the crypto regime, quantitative analysts consistently recommend compressing the RSI lookback period to 7 or 9 periods to increase sensitivity to violent micro-cycles, while simultaneously relying on MACD for broader trend confirmation rather than absolute valuation [cite: 13, 45]. A 2026 academic study targeting algorithmic decision support explicitly for cryptocurrency trading evaluated 625 dynamic weight combinations of technical indicators on major pairs like Ethereum (ETHUSDT) and Solana (SOLUSDT) [cite: 46]. The empirical research proved that isolated, standalone indicators universally fail to produce sustainable alpha in crypto. However, by applying algorithmic weighting to a multi-indicator ensemble containing EMA, RSI, MACD, and Bollinger Bands, the system dramatically improved signal accuracy. On Solana, the integrated approach increased the win rate from a dismal 32.84% (standalone baseline) to an impressive 58.11% [cite: 46]. Furthermore, the method was highly effective in mitigating structural risk, reducing the Maximum Drawdown (MDD) on Ethereum from 50.04% to 41.35%, ultimately resulting in a staggering Return on Investment (ROI) increase of over 2,200% through optimized risk mitigation and the aggressive filtering of false signals [cite: 46].

## Direct Comparison: Performance Metrics Across Signal Types

To systematically evaluate and distill the operational efficiency of these momentum indicators, the following analytical matrix standardizes the performance metrics extracted from the recent quantitative literature across various asset classes, geographies, and signal typologies.

| Indicator | Signal Typology | Asset Class / Market | Evaluation Period | Performance / Accuracy Metric | Key Finding & Limitations |
| :--- | :--- | :--- | :--- | :--- | :--- |
| **RSI** | Overbought / Oversold Reversal | Emerging Equities (LQ45 Index) | Aug 2023 - Jul 2024 | 97% Accuracy (31/32 signals) | Exceptional precision for identifying mean-reversion; generated very few signals, resulting in lower total absolute profit compared to MACD [cite: 40]. |
| **MACD** | Signal Line Crossover | Emerging Equities (LQ45 Index) | Aug 2023 - Jul 2024 | 52% Accuracy (86/166 signals) | High frequency of false signals (whipsaws) due to sensitivity; however, captured massive trend extensions yielding 2.2x the absolute profit of RSI [cite: 40]. |
| **MACD** | Centerline (Zero) Crossover | Emerging Equities (Indian IT Sector) | Apr 2023 - Mar 2024 | 20% to 50% XIRR | Centerline crossovers equaled or outperformed signal line crossovers across major IT stocks, effectively filtering out premature intraday volatility [cite: 41]. |
| **MACD & RSI** | Trend-Momentum Synergy | Emerging Equities (China A-Shares) | 2015 - 2025 | Outperformed standalone metrics | Dynamic parameter adjustment (widening RSI bands during high HVIX) combined with MACD crossovers significantly reduced emotional retail trading errors [cite: 42]. |
| **RSI & MACD** | Standalone Technical Baseline vs LLMs | Commodities (Gold) | 2025 - 2026 | MACD: 17.5% Return (0.82 Sharpe). RSI: 7.1% Return (1.0 Sharpe). | Standalone indicators vastly underperformed LLM-augmented agents utilizing the exact same data (LLMs achieved up to 61.2% return, 2.13 Sharpe) [cite: 39]. |
| **MACD & RSI** | Minute-Level ML Feature Inclusion | Developed Equities (SPY High Frequency) | Apr 2024 - Sep 2024 | MACD: -2.9% Return. RSI: -2.5% Return. | As ML features in ultra-high-frequency data, both indicators degraded the baseline model (-2.4%), suffering from severe out-of-sample overfitting [cite: 11]. |
| **RSI & MACD** | Weighted Algorithm Combination | Cryptocurrency (SOLUSDT / ETHUSDT) | Up to 2026 | +2,222% ROI Increase | Standalone indicators generate massive false signals in crypto; weighted combinations increased win rates from 32% to 58% while slashing maximum drawdowns [cite: 46]. |

The compiled data strictly indicates that no single indicator or fixed parameter possesses universal, cross-asset superiority. The Relative Strength Index excels as a highly selective, high-precision filter for identifying exhaustion points in range-bound markets, protecting capital through rigorous threshold boundaries. The Moving Average Convergence Divergence, inherently delayed by its mathematical reliance on historical smoothing, serves as a superior mechanism for capturing the magnitude of sustained directional trends, albeit at the significant cost of enduring noise and false signals during periods of tight market consolidation.

## Literature Synthesis Matrix: Momentum Indicators in Modern Markets

To provide a cohesive overview of the shifting academic and institutional consensus, the table below synthesizes the thematic focus and key methodological findings of the primary literature utilized throughout this report.

| Source / Entity | Thematic Focus | Asset Class & Geography | Key Methodological Findings & Perspectives |
| :--- | :--- | :--- | :--- |
| **Lund University Event Study (2024)** [cite: 25] | Efficacy of TTRs in mature, developed markets. | Swedish Equities (OMXS30) | Demonstrated that RSI and MACD variations produce highly inconsistent abnormal returns that fail to reliably offset institutional transaction costs, robustly supporting market efficiency in DM. |
| **Robeco Institutional Quant Research** [cite: 4, 5, 35] | Factor timing, momentum integration, transaction friction. | Global & Emerging Equities | Confirms the predictive nature of short-term factor momentum but heavily warns of alpha erosion via turnover costs. Advocates for risk-parity structures and blending quant models with fundamental analysis to elevate Information Ratios. |
| **High-Frequency SPY Study (2024)** [cite: 11] | Machine Learning, High-Frequency Trading, Overfitting risks. | US Equities (S&P 500 ETF) | Revealed that appending RSI and MACD features to Random Forest models for minute-level trading increased dimensionality but actively degraded out-of-sample predictive accuracy, challenging their utility in HFT algorithms. |
| **FINSABER Backtest Framework** [cite: 21] | Bias Mitigation, Rolling Windows, LLM Agents. | General Financial Markets | Established strict empirical protocols to combat data snooping, look-ahead, and survivorship bias by demanding dynamic asset selection and extended evaluation horizons for AI models. |
| **Indonesian LQ45 & Indian IT Analyses (2024/2025)** [cite: 40, 41] | Indicator statistical accuracy vs. absolute profitability. | Emerging Equities (Indonesia, India) | Provided distinct empirical divergence: RSI offers superior accuracy (97%) for mean-reversion, while MACD captures significantly larger aggregate absolute profits despite a markedly higher false positive rate. |
| **Gold Market Multi-Agent LLM Study (2026)** [cite: 39] | LLM architecture vs. standard technical baselines. | Commodities (Gold) | Proved that AI models capable of reasoning contextually over technical inputs vastly outperform standalone rigid execution, achieving Sharpe ratios exceeding 2.0 compared to baseline indicator failures. |
| **Cryptocurrency Indicator Optimization** [cite: 13, 36, 45, 46] | Extreme volatility handling, dynamic indicator weighting. | Cryptocurrencies (BTC, ETH, SOL) | Highlighted the severe danger of standard 14-period RSI parameters in parabolic crypto trends. Demonstrated that algorithmically weighted combinations and LSTM networks drastically outperform standalone EMAs or boundary rules. |

## Conclusion

The empirical evidence and institutional literature surrounding the Relative Strength Index and Moving Average Convergence Divergence present a highly nuanced, regime-dependent narrative. From a rigid, rules-based perspective, the deployment of these standalone momentum indicators in modern, hyper-efficient developed markets frequently yields sub-optimal, risk-adjusted returns that are heavily eroded by transaction friction, slippage, and high-frequency algorithmic noise. However, declaring technical momentum entirely obsolete represents a fundamental misinterpretation of the underlying data. 

In emerging economies, where behavioral biases and retail participation remain structurally pronounced, the indicators maintain significant statistical validity and utility. More critically, the advent of sophisticated machine learning pipelines and large language models has reinvigorated the foundational logic of technical analysis. When stripped of rigid, deterministic execution rules and instead utilized as contextual, weighted features within multi-agent AI frameworks or advanced "quantamental" models, the RSI and MACD continue to provide invaluable, mathematically structured data regarding market velocity, trend exhaustion, and directional momentum. Ultimately, the future of quantitative momentum trading does not lie in the heuristic optimization of the indicators themselves, but rather in the development of sophisticated, bias-resistant architectural models capable of dynamically interpreting them within the broader macroeconomic context.

**Sources:**
1. [poltekkes-malang.ac.id](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGuRVtm9njq_-00HEmn_EVUY0wuPORprbr6AZG0_v31A-RWKl4OQHp0CnIrtmydG3T8K1ZiWZE0kT2-1EkhcL0RE8VLGDVb0mRudKp9z397avBv64aKngsAAUeHcvsiR60Kq__QJBBg-cVB9TpNKlrvS_IzNun4PPYNsu75i0Uu5ic_SeEDwQ4bqOLAwLUtEZdWaw9-syNSDEDQ)
2. [parliament.wa.gov.au](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFD7Lse9uFnWVt18D4rpKlLB5PU47RHEz8fuv9iCl42uQz7O_C9oJ4njJ4udcHZ3fSbDhNNbyGhpdL6nmVtsq-elx9mG7t--tdxAPIMMKV4g0BRVrJ_ndNxpqJw0tXvyWZeE6DjFCH1nycBQBUD-l3_C8_tkFLQBFpHf1RgRHjY)
3. [nrc.gov](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQF0Z9mTkj6BuV2r2ZlQe_Ce-LFQ3J4um1lCTt3WuvZY2sMxNAyYaqWY7xQLs8Aj3uRnJTt5ek2LoNiaaKxj29N_UtjS7fJyKwMB1gVAvd2mhXsOFcBywyV2mEyTXCs2_pSsQxXqBQ==)
4. [robeco.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHfTYm5BqT2GE5Tb-scmee9qw8Lm9YebryvulPs-SO8Ro17Y41UZ-E86ah3riSj4OGRq7hwiqrpZospIONvy1b6wjkvzj6zp9VssKziehbUjxviYkU79VmDsEkFxzMeTp8TOhiDnjjuIF2ygpZ3M64-BElE784ImGcZU7O6PfP9o96ZI8DzIC2bz-1SkQ==)
5. [robeco.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFeUQig8XNNGaFhbVN8eG1qEgOAbcG86TaoS2kCTfs6fHGcc7o34595tHNFR8r6tfiWuCm7P64qAgQCl6C4dvkS6y_1Jcmt8eK8epwWz-pTWlnSrSImqcZ03DmrJ6xfBwzo6uA6zcoO0Qq1szZpXAKXHEymV-lzaqljjSyMicP1R8GfCa9Xs6tjTUppXN2ovjFjQfFROyZ9ZJy5lXmcbd4dGDwEzFG7Yg==)
6. [pm-research.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGoDzuAotE7fjIjLuZCACSvW9vxiWEBc6oN4enzosOpCF36rsZeeVovAja1CF0DYaEWE7zEEoFcuCEsRVYXT3ceIU0AP20zdvmitWq3Zi0avoEBmdOD-rjIuPvX2MIn9LNRr-EwpLBFmyvUz_g6WYyYnEdrulBAk4KCoUXNh8yTNDxA)
7. [9823capital.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGlS2eEFN0fQHpRe_9mn6LDwldt2l9ESMp55MhXgB7kh5xk_0XMsCwYHX6Q009WSHzaHi1jMJkp4FJIgGFMCb26sDB-U2gklDyBZft_YYU7iwN_p_CLum6zUuejrYmSMMjHXqzJgwEkx-lelbiN7bRpmqcHWo5v9VrTMCbrzf7y3iC4KVNCXQ0ZsNd_QBY91NAlO2zK_39C302xor8ncygu-CKUMA8=)
8. [substack.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQESBJCIeumtDE5itYlnRQDsah0NtyCKDOaMoDsbgi4SyeVhKvd7aBAqujyoSTGZ9Af4ZzeD3JlzMv23o__np9nK1KXW9bIiVK4jF_kmVZ_gpHNWKCwjwCSnbg99LiWk_7RbNVcl1M_wMEIXOj-ObEtY26cL1K0Xa-A-REA=)
9. [wikipedia.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHTi2BFHLOhVu2OV7QLHAMutJWlUdumRj0WUB-uYjVi3VhRLWlezNsWPslw6hXash0JiXSxkpUYDlpuW8Wxr3Curf6w-EEsG6n2pCFNdyTVqoGA5qbeEfcNHBGxGYHNHpJ7juPZ21D8q6wXvw==)
10. [fidelity.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHJcmnibDJFnmx2T22YIPCnw434st-I1JVuPqoFKOC14ONmq1NCeZKh2Jnbq0A8QTp91WEpeIdLDpa6E1wu9Rx2qxjF6xQgye0DtAft-BChU6ZfGiB-2f0-BO3RTm3_YNyVKGx31zKrRtZ-lT7nG_Kynq93zDmfIWE8DMzvOkdiuj5cnXKes2YSF9M-AD1onmM92EsyupVJ8QBoa0M8lzQGag==)
11. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHXLd-iZG6pcmTqGg-lfIJny0Yb8lEEx9MkX5-mMTp-eR3EorC04judUcwQM92vx7HocYN7gT-B_axFO7anGh5xhzUEdIjHq6zRStqLL0cyEK8PCtA9qX2P8A==)
12. [gtcfx.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQExDPwYHkLOJr7HezTQxRfHwgnoejQF5TzQsAwVyvHj574PszCKnD1JNz78YiLGVPHJnxD7pKqgOniPl-MtK6JEcWbfo9qaxsEQimJv7gZRn_4A8wrr5NQNLqj6dOWlpUBTAf871Vn6hmxEjmnDh_G7)
13. [kavout.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQF3E3VkRyxUK_yv_Pybp6EvjFmdt1IGHCv0ktqQlywN8147BLc8KHzWxQdhUTFXFbwxK5l5F-hqS1ZphYcJs3_-fOfbv7W2xKFY9YSUQkNskCfYOAATlCrgpiDWpqh4Hg2yN9KsnvIsaXFhxNj8aLCQgjtSjljVHwwOPOlhnw_q9R5yanR2SlQE1-Ru09EdS5tWmEi2TSSJBHKIqK_Ck5EBdU0=)
14. [tradethepool.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHoORt0siuuyLI94qhs8KBH3utySFrjd1ClvjlfhtVNN6fyIg6PErRvCfvVz7d1njNcCIR5LwLYHZTrbHgaur5NTqVUk9evtQS3VYx2KGXPbYYN-8hJjpn6iLympsLMU7z4Sn73iCrCBCr1Bg==)
15. [wealthsimple.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEaK3wTwhFO3V0-Dyg2GN-JSof6a6a9MeeThs0hDnoKH6rOZwbLHFvk3vdkj_GH3TM9sW6-lpgMpLIPYlWuuuQyE7364JVdYO43MMpd6WbSfXTKMajVyQS31dDk77gwwhLyM1P3jex80H7pxg==)
16. [frontiersin.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQE61PlHvPUh9EY7J39DTLQiP53gYoL-WZiBmlTIrdVbgeqTrKRzkIs_ZeaNHKzOc5F9MMFzMAD8S6p2ewoSPKo0qsjlpBG03V2Xs-8tPZSgzKI9_ImgdTUMARG3hOCvukyJ8wAU0RDyAMyeGEFhJ9-zysPMEwmWvvwqAbiK9VyuSHPSNBKN28FdETi9IA==)
17. [b2broker.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHhc8q6teuH-tVWqnAjdjVgJnfhIdd3TaET6lwN-U2b9K67CrFB2RntLMUlrcvyyVsCdYrvM4GK9ZNtgqS_pAgJzscpSahj06KZ13rjCQw2TFz1wZa5pQ9eaUnZBH9MAZFX3a39i3HGrn6IAckOsPI6aJl9om8zqWLiQ-x_9bp7r-2O0k-F7DVpGb_rjKqoX6g=)
18. [Link](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGg3K88topRQMS9I9cgofJh2ivxVMew5JIIzfaqZ-Y3B1kWBPjpG6H4GX1pFwnWF72u65xKNramkecJE7nJdzFN32uCPPlOwcsGbFTg63UvvHfBv7xCS7bPKpo=)
19. [litefinance.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFb5m54TugBJuEijtt4_CReMnA55WEkwjcMPt-dEL7t4zT1T2PyNR0vlWSKO5hHiM32en48yjYWxbiWwoeTQLvBbA7QkgpvZypFFzN7gNDcYsjOKVWJtpy-c_pzX5giHVWAYRcQkWhP7f4HBdlHD0MmVNhmXvCCIZFRvmswym78o8OyYsIPb6va4GLy)
20. [ecommerceparadise.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQE9GBtIuDdHmvv0eXWj1Ke7qJ4tQ5UvlWvS4YRmqX1YW011eXtaW6vE_ZFbNEGRSxgTMHiM57yl9OEK0pYYIHl1s3MG_mPzuzqIEtwiZlIaq7C7vcVr05-5GMmMGg2CmrzA1-1nGDgO-ZzarLCk4kEHs_YvEzUueEAy2UaFFM4nk1fPJWudkp4IpztvoVB8W3t2sULbPmGBr9SV_C5N1j5cUAegqavTRpw_8mL7FJsjZYoD-VZ6)
21. [arenafi.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGV4qZWDuBRmT2O4prsFTwxY0gbwhJko_z33gpUSHKeTLoQjXJvkhvWLJd_s1IAIhNJOkskE2yZ_f7Vya9yHDKgr2DnIR60af7keVciMijSk1bmVRPUfy5D-gURDIffa205tBE4vrI2PcGQCg==)
22. [emerald.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQH5nMPNJ7eOfc3gMuNpBdGsfHDp_5XUU7LEg2JtkuRRLZ5TQ55DjOj9r_0_qh8PW_0k1aMPoVC0Z3X8tU-DYGBh4sFE1ABj6h2wc_JXzaVK9WURhnGBS6yBPV_JWWtjjPAKYS81IE4WFZOAgTWfvagfkKIcrlTM2R9ERDRByYQ5pKvDa9_1n5we9EuX5q4m4g9g-3zre22-ZWwEkw==)
23. [ijcsrr.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQETxGhfNaWWoZL9elgLz9uvjIFPRE0mBFl5s4WpFYgXMMkDkENGeM0F0nuY-ALu80hwhrNQHnxg-LvjZgc3eB116qrFH-wTbxwWqyJC8vBmoVqzgLHGh1eJ_R3vsAVLzAM4DuTI9wIkkXopv1wC05fIC46MiQ==)
24. [shirazu.ac.ir](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEZhxr6HCfHJG0bstU1M8UwV7WYkFaI_RjKC3XwN-iOOHugVq9Dk4pi1taBVL4Udhj5czOoQUJN1W3dqhqsN5SAVoPoDgMDNld81tN7RFwgu6TvujzGiJfaZtdbwWI9xAtv4_wxL6vXt3x6Qbn4HC-WEmNtBZeBGQ-x38OUTbi5v6IA)
25. [lu.se](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEh3gsHJHVO4hUSGz_5BeAdIHIt4ie26hdOSTn7DWx1pSYac2FQDtcooyBi0dBHoLNw2_IoTl8oiZDj3RsKyc99Wzwf1p6hRah-OQkgH7HG5esFhUZ2VEDslG5CEBYPq8mWFaZ4PST-5ZT1Qpspnu4TBOTPTJaVwuvOOw==)
26. [mccrackenalliance.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHXnzGstXH9nr0JBnpNTv7jdl3RZB5d4LsM2DlMOCewXPPBdyasYyyS43KCjGos9fWIP0UyYJRLScstH34QePc4yKXCEXlmEaZ9l6bSswGEmCCaLgOqy42oef936SajHeU_VQqQENx19VkkX1Plq_RbRJVz19dm-V9WG2nSxww-a3_S4oh46q_jtiGChRgLst8Se7UtTGpYgDM_qMKL5v35phClxeeGDcKH)
27. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEvQyDkcoDw6VURtYLbuDVK1hMILKW69aEX-pSUbsnwsNlchtdpVDCTatBS-9Qkk3LyPri-wdguREAVuDPNUM6asUozfZoFiV0oddHjACCY5DFlpMeKnsY9Hgef0WdDGAHwUxPidfp0daj5MAguS5eyojsFf4C7oPSGJRLi_jQaYQJtPUaVjRFcZx5BHEUP82YeI1JQ47TDW-zP3LDvghcVSoXAI2J2TX-fLGlXZVOLeunkZwc=)
28. [preprints.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEhua8iGNrVnKDN8hEavDqO6W3ECGIH5xG5fMoCeZFkxLnON_KvcDMF3dU2oam36AEd-uGI8b96l1UyhjkjuI4L8neY4mPK4fVThVjRdE-Grtx7rxnRMi1v0_Q6UJqaLkuN7MmPSQw=)
29. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFEtywTZKVQ4aJnc1iqQxaI20rfBDsPD2NwPzvIk86W2bThsUlJVsTUsBmxI0kyg63jwleb4MM9Qc3v9CZM271q1yM39vY-ENhuWRIQLJCUiKu6oyOOd2U3itgpuN-VkSVFlbpMYSKMRmW2m7keptzFHmnPRuBCg9Xl_ggBBilE7RWKb2Bv9dc3aK6Br-hVjtGiMPXc-4bmcKZsnSXwn5YEQVa-Lvd_vytkGACizMfP4YipMJeEP5tsQbo6nBuNhMFLStL9y3k3WY2N1AbZoBluaW7ftCjRvFXd7qXJJJMQ3nqVPv_NVlx6fydVw6o=)
30. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFl4-K25wS5VAuYdPny1yUUiFzLMEHMGlYZL4CvOO-hT0m1wNT1fBeGmztarSaNBPEthGMRAHB92oFQ4XefNlgv2VAeKjOEIkE5Ries9rklFyjiWBxspXjevQ5aiQ21tim-vqk7cTURw_1-rilsVtFhISkd5kJhIljz2lJXswDMeO4tivxF5yQaYGDXxLlAbOMh8OthgdEdyA4WrOKHQkC182gBaE_caBL3GWaqcR9xCgdac8nd9JtWw2vp10akZs9b-usJ7Cu-Kw4HurEFKAzQvK9KynQ=)
31. [instatrade.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEy1ZMy7xEIDKFMF7RfVhmCFS16TaCN2csHjoco2eNtIYSSzuFGJWQI0fNeWajqtBYgBuIvrLxCkKmDEDZhlDGHxzJN_6cJXGunGjRcizz6pgoXcgm2tvMWD58CuFNZLHcIkEBOKDf05VQ=)
32. [medium.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFCJFAHSvXSb_Jvt8spKWx6mMkXv-KedKSRcyCkPnerqlVLI8-eHxUoNYtLWlFJhKgTDF2PZvKEdNTa_JqFPZa_GImzW8cpfEAbIeIZI3wE9sKVHPlVZFlMD4atsYoHQilFJT7eBbc2_HOKLX42Y8Kpd7q0UGxzZHmVqIL5-7sCydr5VRcClrxaD_SmpItd58OQ1EulCPwYCQP4ClMMPictyN1E8ef6NOGPOToznB68WQ==)
33. [pm-research.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQG58AUh91AdFIfPffX0RcTamRnII2zldUPS4FhMR-T7kXU0zDtJOxmd5faxVEJIJInQ0bkw0Yu9r9AbgBAePYLxV1fdxCuzSk_pjgeAC8hNkVcl4q-_k187mztrcXtIqih8odq-DGJIeZRaKuMBx1Qma5c-BqaMwPXb9QbONVg=)
34. [robeco.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGTqx5_FRSHp5b-qk8UUiDq-SBQJkfsv7oDwN3XOtaV51TmIbviPDJ1hwmOiuUcDBKc6wQAq64ZOsm7LzFzMq2F5v8xpp9YcJYEm1x6jOcLf3ZqhpHZoxrok9xTYJNodGegYLOv2WwjxswvZ0vtiv6Lm9ltxAofWbBlmAdbZPVWnI3zDgI8tyuEl270507vGxxzLF_58hfIPS8ZikpKRxqLFwdaC7vuRQ==)
35. [robeco.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGCpNG_6PJlYp8K7dyeP5LomUFfKMBSz1lBbFU0QFaUFWz44ShzTw_bA-9nN1gGCrgAJu7t1O45_lqzbEt7z7hzq6uXUAExbuH0sTkpSbSXIHhqDIt-GII9PjqjZBzFoy4guoILhmt7O2Qoe-tAQz_EGy-l1iYp9O655PrJyvqaE7TPyQN1wKIZnK5FQxd5tmMv-V6-20hSotiR-jRy23Xx8dblT4jzgcRAgvhIXu8=)
36. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHhhekZKLaMiQIslTi3Spzej1fUl-tGE-T4dS57uvhJT4Nrj9o8p-eBM3SzkYx2nkacZqMK5CtZ7sBiXPfgkis3epc51exWcOQPSB6C46qp5zSt_vezSQ==)
37. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEetKujnBALqizJuDcDF392EP6amSkcxW6kzyrP7q1-svkk3sNMed2ePP-NSzTseMZvL8F5UJWze4OLhRKSZKfCS7qL_7XCHCqGQ8HTyYfsaMdnJRJ0Zf2teQ==)
38. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGpVHcTvPWaUZ0FeagEC1re1MfqAKg0AbrgjXpqUFzezV-Bs61lgcyGealTiX_2fz7gwNQPhjn7apXQEstK01o9XyO7qdTCTQcgDRvkovkgdfoWn21SGVTsBg==)
39. [diva-portal.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGK1SL0LDqvpt7Vm13CgXjRVAfBFXKTRF0aoS5z-RkTN1J1X8T-x-SuusmxAHuLVd_3fyOfxEauPzYW8k9JJ3F_sMjOApvsf84Bn-FMdiygf6rqSv4mpk4qDmvN0uNe-RnYNVbwQGQ0BjFXO4H-ma5zRviyL-oa6w==)
40. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGTTnuW1g0c-2QyCUCExKM8d-VyD0lj2wl5Tz9GRI2WMMeZeZ-2VK8v0AzO5eFb632P_E7hac1V6PGs-UOPcI1Xl38kxFBOcbnKc2dQE-e1TdWEDin2YKFNebh99whfcgMMUtSK-ll5oBJAJUFtFr0Cpr-1hM3GQbaUACpzJpNg1mzJqnjyOTQy7D9Ex9JoA_KT50zeaMMCxzmodUeppdCqpC44Bxaf0WZClAl-F9bkehLtpZdKHMitxgd-K4yMdn_B87jxLSdP)
41. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHgjj46hofX-VJaieCpuIT64u60JjU7RU9RbovJOiKIWwx4KrHfQAHhicWMzzDYZdodt-Gd3Dec85t2eCrjSbfYCbxVrvl3zvq3Q9F5-18b0HDg19P3n7K4Me-FiHvJZppbydqQO-akh8CuiyuoUqCStFFx4S8W_XxxVBCN-cYb7H7p2GtvL8hE0kSDFLdVLIKe2Zli43IVdvdtB4co0aTLwleMqlr-K8jZ_vqkYMAnkiqKhSLc5-Z_)
42. [Link](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHFC3RBSmYwm37tKw_iX4TPzYyJlNPuDiVnkQdBsIVbAM9Fnm5GojJ1A0Z2eBdMo1aCnbwlEvXYygnPWqNfWFt9pdRm8O8zZzTnMAH63vKhTdAFirXUhvuvSfzYpWAdv3IAMzEaP2E3LVl3QERz9sPR4zXVVkqwqpgdooZJZI6DAQ8GQt5r-k79LScR)
43. [srdsjournal.eu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHW9A9D7OXn-g2vOa0GOs8XI5F4MApWQ6HUuW-trTLmMMbiKkz3Gvx4668RqJblFiTVYLtAMuotrDzI_tygwmvxk4yiGuIjv2BfX4Vldslvxbahgno75PS8hUocVN1W7ozDCtRqbwYGoj0OkcU=)
44. [researchbank.ac.nz](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFSxi5QnDL6lmvLk1VJhKKecSMqc-C6bT94y3zooT4kcDQbojsYb2ZETRmwBNyALWD975NbZ_Fux0k4FTXq2eACZwxNWoAy4JRFNBSIZD4kKq0Te-RJNtEOqLPx7rCp-9w1bIpJ6PGsVEOjDMXTGCQCOFadCjGhMjUn6VwIq1yAO7-5sNTl3ZNOfDmTvT8=)
45. [altrady.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEZ6NOMJaj0nOFzx4h6_Ud-u1_deBQl2aWV5PTJAgss_GF_2Td4gcIdrRwy9wmD2_FuEbAREanrgO7u8QA63Vf0wjidx1ZCX_m8ikOkBHIj27Ph31c1eCl9vGf-GLzwkkNPxgS5-kcG5BxTM20Wn8YYs3UlElkHVQypDiS6wJC97U08SQhIN9sV86m_lSIv)
46. [dinus.ac.id](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGA8zNARhhl4esayTvKbOzAiIA1_ZOMJ5YdoCImJOT1E0OagELJI0XNF8Z0-J6lUE3U07yqEMoo6RjDojnIk08JcBpjEtoIP3UO2boc5jYegFERWKMZiGztQ1l5oa4di0gLG5R1o_QSrYtvNL0=)