The Oxford Alpha Fund (OAF) deck presents a quantitative framework for enhancing traditional momentum and mean reversion strategies through the use of volume filters. The core thesis, detailed on slide 3, posits that price movements with low volume (normalised volume < 3) indicate momentum, while high volume spikes (> 3 standard deviations) signal mean reversion due to market overreaction. The presentation relies heavily on backtesting data from 50 S&P stocks, demonstrating that volume filters significantly reduce the total number of trades—from 3.64 million to 2.67 million for momentum—there…
Key takeaways
- The strategy uses a 1-year rolling basis (252 trading days) to calculate mean and standard deviations for price and volume (Slide 3).
- A volume filter of > 3 standard deviations is used to trigger mean reversion strategies, assuming high volume indicates panic or overreaction (Slide 3).
- Backtesting on 50 S&P stocks shows the volume filter reduced simple mean reversion trades from 3.5 million to 795,848 (Slide 5).
- The deck uses Microsoft (MSFT) as a case study, showing a price increase on Nov 10, 2022, with moderate volume (1.5σ) as a momentum signal (Slide 7).
- Future expansions include using Random Forest (RF) and LSTM models, which achieved higher test accuracy than random guessing in classifying price movements (Slide 9).
- The presentation lacks a team slide, financial ask, or company roadmap, focusing entirely on the mathematical strategy (Omitted).
- Data research indicates a sharp decay in the number of samples as normalised volume increases beyond 3 standard deviations (Slide 13).
Oxford Alpha Fund: Quantitative Strategy Teardown
The Oxford Alpha Fund (OAF) presentation, titled 'Augmenting Momentum and Mean Reversion Strategies with Volume Filters,' is a highly technical deck focused on quantitative finance. Unlike a standard SaaS or consumer tech pitch, this deck is built around a specific hypothesis: that trading volume is the key variable in determining whether a price spike will continue (momentum) or reverse (mean reversion). The authors, L. Tan, D. Seiferth, K. Al-Robaie, and Y. Jaiswal, present a data-driven argument for a filtered trading approach.
Slide 1: Title Slide
The cover slide establishes the academic and professional tone of the presentation. It lists the four contributors and clearly states the objective: augmenting existing strategies with volume filters. The OAF logo is prominent, but there is no mission statement or 'elevator pitch' included here. The focus is immediately placed on the methodology rather than the brand.
Slide 3: Introduction to the Strategy
Slide 3 defines the core logic of the OAF approach. It breaks down the two primary market beliefs: momentum (price movements continue) and mean reversion (price movements revert). The 'Goal' stated is to find criteria to evaluate which outcome is more likely. The team posits that price movements with exceptionally large volume are likely overreactions (panics), whereas moderate volume movements are driven by informed investors. The specific parameters are defined: large price movements are > 2 standard deviations from 0. The volume filter is binary: a momentum strategy is applied if normalised volume is 3. All calculations are performed on a 1-year rolling basis of 252 trading days.
Slide 5: Backtest Performance
This slide provides the empirical evidence for the strategy. The backtest involves 50 stocks from the S&P 500 with a 1-day holding period. The slide compares 'Simple' strategies against 'Volume Filtered' strategies. The data shows a significant reduction in trade volume. Simple Momentum resulted in 3,635,402 trades, while the filtered version reduced this to 2,674,746. The most dramatic change is seen in Mean Reversion, where the volume filter reduced trades from 3,500,057 to 795,848. The charts track 'Cumulative trades ($MM)' from 2020-07 to 2024-07. While the reduction in trades is clear, the slide does not explicitly state the Alpha or Sharpe ratio improvements, focusing instead on the efficiency of the filter in reducing trade frequency.
Slide 7: Volume Filter Case Study (MSFT)
To illustrate the theory in a real-world context, slide 7 examines Microsoft Corporation (MSFT) on November 10, 2022. The data table shows a price increase with 'moderate volume' of approximately 1.5σ (46,268,000 vs a mean of 31,000,000). Because the volume was below the 3σ threshold, the strategy classified this as a momentum signal. The slide includes a screenshot from BleepingComputer regarding a Microsoft Patch Tuesday update as a 'Possible cause' for the informed investor activity. This slide serves to ground the abstract math in fundamental market events.
Slide 9: Future Expansions and Machine Learning
Slide 9 transitions from static filters to predictive modeling. It outlines five areas for expansion: cross-validation with ML models, generalizability across different markets, varying trade lengths, investigating more parameters, and predicting movements using volume. Four charts show the performance of different models. In 'Classifying (in)significant price movements,' the Random Forest (RF) model outperforms Random Guessing and LSTM, achieving nearly 60% test accuracy. Similar results are shown for 'Cross Validation of strategy using ML,' where RF and XGBoost both significantly outperform the baseline. The right-hand charts show normalized volume spikes at 'Time 0' for both daily and intraday movements, suggesting the model's ability to pinpoint the exact moment of high-volume price changes.
Slide 11: Extra Slide: Past vs. Recent Trends
This slide provides a deeper look at the distribution of observations over 3-year (3642 observations) and 5-year (5544 observations) periods. The bar charts show a massive central spike at the zero point, with a distribution of returns or volume deviations around it. The 5-year chart shows slightly more variance in the 'tails' compared to the 3-year chart. This slide appears to be intended for technical due diligence, proving the stability of the data distribution over longer time horizons.
Slide 13: Data and Research
The final slide in the provided set visualizes the 'Cumulative number of samples' and 'Number of samples' relative to normalised volume. The charts show a steep power-law decay. As normalised volume moves from 0 to 7, the number of samples drops from nearly 400 to near zero. This validates why the 3σ threshold is significant: it captures the 'fat tails' of the distribution where the most extreme (and potentially mean-reverting) events occur. The slide is dated 5/5/24 and notes that figures are in $MM unless otherwise stated.
What OAF Does Well
The Oxford Alpha Fund deck is an excellent example of a technical strategy brief. It avoids the fluff common in early-stage decks and focuses entirely on the 'How' and the 'Why' of the trading logic. The use of standard deviation as a filtering mechanism is clearly explained and visually represented through the red and blue arrow graphics. By providing a specific case study (MSFT), the founders bridge the gap between quantitative theory and qualitative market news. Furthermore, the inclusion of backtesting data across 50 stocks provides a level of statistical significance that is often missing from speculative fund pitches. The forward-looking machine learning slide shows that the team is thinking about the evolution of the strategy beyond simple static filters.
What is Missing from the Deck
While the strategy is sound, the presentation is incomplete as a 'Pitch Deck.' The most glaring omission is a Team Slide . Investors in the quantitative space bet on the pedigree of the mathematicians and developers; without bios or credentials for Tan, Seiferth, Al-Robaie, and Jaiswal, the 'Human Capital' element is zero. There is also no Ask . It is unclear if they are looking for $1M in seed capital to build the infrastructure or $100M in AUM to trade. Additionally, the deck lacks Risk Disclosure . Quantitative strategies often suffer from 'overfitting' to historical data, and the deck does not address how the strategy would handle black swan events or regime changes in market volatility. Finally, there is no Business Model —it is unclear if this is a hedge fund, a licensed software tool, or a proprietary trading desk.
What a Founder Should Copy
Founders building technical or data-heavy products should emulate OAF’s Case Study approach . Slide 7 takes a complex mathematical concept and makes it tangible by showing a specific date, a specific stock, and a specific news event. This makes the 'black box' of the algorithm feel transparent. Additionally, the Noise Reduction argument on slide 5 is very compelling. Instead of just saying 'our strategy makes money,' they show that their strategy makes the process more efficient by cutting out millions of low-probability trades. This focus on 'efficiency' and 'signal-to-noise ratio' is a powerful way to frame a value proposition for sophisticated investors who are wary of high-churn strategies.
Frequently asked questions
- What is the core mathematical trigger for the OAF strategy?
- The strategy identifies large price movements defined as > 2 standard deviations from zero. It then applies a volume filter: if the normalised volume is less than 3 standard deviations, it follows a momentum strategy. If the volume is greater than 3 standard deviations, it triggers a mean reversion strategy. This is based on the theory that extreme volume signifies market overreaction.
- How does the volume filter impact the frequency of trading?
- According to slide 5, the volume filter acts as a significant noise reduction tool. In the backtest of 50 S&P stocks, simple momentum trades dropped from 3,640,000 to 2,670,000. The impact was even more pronounced in mean reversion, where trades were reduced from 3,500,000 to 795,848, suggesting a much more selective entry criteria.
- What specific technologies are proposed for future strategy expansions?
- Slide 9 outlines the integration of machine learning models. Specifically, the team has tested Long Short-Term Memory (LSTM) networks, XGBoost, and Random Forest (RF) models. The RF model appears to show the highest test accuracy (approaching 60%) for classifying significant price movements compared to random guessing.
- Is this a pitch for a startup or a hedge fund strategy?
- The deck is titled as an 'Investor Presentation,' but its content is structured as a quantitative research paper. It lacks typical startup elements like a 'Problem' slide, 'Market Size,' or 'Revenue Model.' It is best described as a technical strategy proposal for a quantitative fund or a trading desk.
- What timeframes and assets were used for the backtesting data?
- The backtesting utilized a sample of 50 stocks from the S&P 500. The data shown in the charts on slide 5 covers a period from approximately July 2020 to early 2024. Positions in these backtests were held for a fixed duration of one day.
