The Risks of Overfitting: A Closer Look at Pair Trading and Cointegration in Backtest Results

systematicls @ twitter

orig

I'm seeing a lot of excitement around "pair trading," but it makes me groan. The usual buzzwords—cointegration and ADF tests—are there, and of course, there's a backtest making a case for a very specific pair, ETC-RLC, out of hundreds of combinations. This is classic overfitting, yet it remains the go-to signal for many quants. It's funny; I lost my quant virginity to pairs trading as well, so I get the allure. Just don't get fooled by the shiny metrics and be cautious about validation.

@quant_feed

Recent Enhancements to Data Infrastructure and Trading Tools

BeatzXBT @ twitter

orig

- Implemented parallel data feeds; integration of websockets proved more complex than expected. I explored various options for managing client sessions and tested ZMQ handling by running 50 identical Binance full market depth streams, achieving an impressive 100K messages in ~2.36s.

- Started further normalization of order types and sides; considered using dataclasses for simplicity in trading logic and OMS, but faced performance issues with serialization and deserialization.

- Divided features into categories for easier updating. Deleted existing strategy from the SMM and decided to rewrite it from scratch to incorporate new insights.

- Transitioning from a skewed pricing system to a fair pricing model to enhance reliability for features.

- Tackling the design of the risk engine, focusing on the calculation and structuring of global, exchange, and symbol deltas. Encountered challenges with JIT class construction in Numba, leading me to revert to native Python for better manageability.

- Looking to enhance exchange integrations while updating my existing HL code, and actively engaging with peers who might find this progress relevant.

@quant_feed

Exploring the Impact of Tails on Crypto Data with Toraniko's Factor Momentum Analysis

0xLightcycle @ twitter

orig

I experimented with @0xfdf's toraniko GitHub, incorporating crypto data, and uncovered that crypto datasets exhibit significantly longer tails than traditional equities. The exploration included three plots analyzing the toraniko factor_mom across different universes: utilizing the full crypto universe, a dynamic universe based on volume, and applying higher clipping through winsorization. These insights are crucial for understanding risk and return profiles in crypto assets. The framework presents a robust multi-factor equity risk model tailored for quantitative trading in this volatile space.

@quant_feed

Rethinking Model Evaluation: Why ROC AUC Falls Short in Business Contexts

predict_addict @ twitter

orig

ROC AUC is fundamentally flawed as a metric for business applications. It lacks the nuance required for actionable insights, particularly around customer churn predictions. In dialogues between data scientists and business managers, it's crucial to focus on metrics that drive decisions rather than abstract performance scores.

The emphasis should be on understanding the model's utility in real-world scenarios, not just its statistical accuracy. A tailored approach is necessary; employing thresholds that cater to individual customer profiles could yield more relevant predictions. Ultimately, aligning metrics with business objectives is vital to ensure our efforts translate into tangible outcomes.

@quant_feed

Enhancing Sequence Processing with Transformer Memory Integration

HighFreqAsuka @ twitter

orig

The approach involves breaking long sequences into shorter segments while utilizing a transformer for each segment along with a memory vector, which helps to maintain context across segments. This method improves memory retention, but the fixed size of the memory vector at 10 raises questions about its influence on memorization capacity—a variable that deserves further exploration. I think curriculum learning is a significant factor here, yet it wasn't ablated in the study. Additionally, this method reintroduces the sequential training drawbacks of RNNs that Transformers were designed to overcome, affecting GPU utilization. They cleverly mitigate this by leveraging a pretrained model. It's intriguing that there are two separate studies on this; one likely overlooked critical observations that the other picks up on.

@quant_feed

Rethinking Quant Skills: The Role of Analytics in Portfolio Management and Risk Optimization

systematicls @ twitter

orig

Analytics and optimization are the essence of quant skills. Portfolio managers may possess advanced capabilities, but much of their skillset diverges from quant methodologies. Enterprise risk managers embody the true quant spirit, adeptly managing risk through analytical frameworks. The humor in the hiring process highlights a key misunderstanding: it’s not just about PhDs—it's about finding minds that can digest complex data sets like Compustat with precision and attention to detail.

@quant_feed

Navigating Coffee Quality: An Australian's Travel Challenge in Switzerland

liquiditygoblin @ twitter

orig

The biggest hassle of traveling as an Australian is lugging coffee gear everywhere because the local brews rarely measure up. Swiss coffee is especially disappointing—it's like drinking flavored ash. Also, I’d rather avoid chains like Starbucks altogether. Compared to Australia, other countries just can’t compete on coffee quality, and even when skiing in Japan, their coffee is just average. When you’re shelling out 5-6 AUD for good specialty coffee back home, it stings to pay about the same in Switzerland for subpar options. Plus, the daily cost to ski in Cervinia is outrageous—totally not worth it.

@quant_feed

Understanding US Bonds: Transforming Debt into Liquidity and Investment Options

perfiliev @ twitter

orig

US bonds represent money rather than just debt. They offer high liquidity, allowing you to sell them easily for cash. Alternatively, bonds can serve as valuable collateral for borrowing. You can also choose to invest exclusively in bills and wait for maturity to access funds. This versatility highlights the functional role of bonds in financial transactions.

@quant_feed

The Pragmatic Approach to Market Insights: Balancing Quantitative Models and Real-Time Trading

liquiditygoblin @ twitter

orig

Price action reveals market sentiment, and understanding this is crucial. Small lots can provide insights into trading behavior without the burden of extensive modeling. Focus on efficiency; it’s about smart execution rather than exhaustive analysis. The goal isn’t accolades—it’s profitability. Leverage quick, tactical trades to align with market psychology. Prioritize getting a feel for the market over complex methodologies.

@quant_feed

Rethinking Options Strategies: The Risks of Trading on Personal Price Bias

salr_nyc @ twitter

orig

Market-making is about setting the price of leverage, distinct from typical trading strategies. I’ve seen effective trading groups, particularly ex-Salomon players in London, who excel at understanding the value of risk transfer and are willing to pay for it. They appreciate the intricacies of pricing risk, which leads to better outcomes. On the other hand, I also have a solid working relationship with a prominent firm that thrives through cooperation and mutual benefit. Their collaborative approach has yielded better results for all involved. This highlights the importance of understanding not just the mechanics of trading, but also the dynamics of working with different groups in the market.

@quant_feed

Embracing the Basics: The Value of Starting from the Fundamentals in Quant Finance

quantymacro @ twitter

orig

I find value in embracing the basics; they’re the foundation for deeper understanding. Critics labeling my insights as "too basic" miss the point—it's a signal of growth potential. Sticking to fundamentals at this stage sets me up to tackle more complex topics down the line. If I'm already advanced and still presenting basic concepts, that would indeed be concerning. Focusing on low-hanging fruit allows me to steadily build my knowledge and skills without feeling overwhelmed. It's a strategic approach to learning and mastering more intricate ideas as I progress.

@quant_feed

Leveraging Rust in Delta 1 MM System: A Timely Production Rollout Amidst Market Challenges

0xLoris @ twitter

orig

We're officially integrating Rust into our Delta 1 mm system after a year of pushing through the bear market. We deployed at 3:50 PM on a Friday before a long weekend. Big thanks to the team for their hard work to make this happen. Zig looks promising, while Redis isn’t hindering our performance. We primarily use Python since network, matching engine, and exchange optimizations dominate our focus—previously, reducing latency by a millisecond wasn't our priority.

@quant_feed

Enhancing Mean Reversion Strategies: Adjusting Lookback Periods and Volatility Thresholds

GoshawkTrades @ twitter

orig

Mean reversion strategies can falter during sudden volatility spikes. A straightforward adjustment, such as shortening the lookback period or setting a minimum volatility threshold, can enhance performance significantly. This tweak has been beneficial for improving various strategies. We've assisted over 200 traders in refining their approaches into systematic strategies. For those facing challenges with execution or seeking quant development support, there are solutions available to help automate trading effectively.

@quant_feed

Exploring the Top Craft Beers for Quantitative Analysis in Flavor Profiles

therobotjames @ twitter

orig

Garage beers and shed beers dominate the craft beer conversation right now—both are praised for their quality and distinct character. The community is unanimously on board, with shed beers often considered their close cousins. There's a clear consensus that trying these beers can be a transformative experience, with some even calling certain brands the epitome of luxury in the beer world. Enthusiastic endorsements highlight their exceptional flavors and suggest they stand out among craft options.

@quant_feed

Exploring the Viability of Low Win Rate Strategies in High Frequency Trading

larpcapitalwc @ twitter

orig

There’s substantial potential for decent risk premia in low win rate strategies, provided you execute a high volume of trades. Picture holding a strategy with a mere 10% win rate and only one trade per day; it’s a psychological challenge. After enduring 20 consecutive losses, maintaining confidence to keep trading can be daunting. This highlights the necessity of trade frequency to capture the risk premium—low win rates can still be profitable if the number of trades compensates for the losses. Emphasizing mindset and resilience is crucial; traders must prepare for long losing streaks while trusting their system.

@quant_feed

Understanding Assumptions and Best Practices for Candlestick Close Prices

crypto_hades @ twitter

orig

1. When calculating candlestick close prices, relying solely on the last trade price can lead to inaccuracies, especially in low liquidity environments like trading shitcoins.
2. Consider alternatives such as midprice, microprice, VWAP, or TWAP for a more accurate representation of close prices, as outdated last trade prices can skew results.
3. Accurate close prices are crucial if you're developing mean reverting strategies on short time frames, like 1-minute candles, where last trade prices are essentially worthless.
4. The main goal should be to ensure that your researched signals are executable at reasonable costs. However, using advanced aggregation methods like microprice necessitates extensive order book data and robust infrastructure that's often impractical for bar-based research.
5. To determine if your strategy is viable, implement a close-to-fill measurement that compares your computed close price with the volume-weighted average fill price upon trade execution.
6. This method allows you to effectively track discrepancies and supports backtesting by providing real execution data rather than arbitrary assumptions for slippage.
7. While similar to markouts, close-to-fill measurements are essential for non-continuous signals, especially when there's time lag between close and execution, as price movements can be significant during that delay.
8. Ultimately, combining close-to-fill analysis with markouts helps identify optimization opportunities in execution, including tracking close-to-signal latency.

@quant_feed

Enhancing Feature Diversity: Insights from Early Quantitative Experimentation

BeatzXBT @ twitter

orig

Diving into advanced feature engineering, I've leveraged new datasets to diversify my models, though the performance hasn’t yet yielded significant alpha. Shared code for reproducibility; though it's pretty basic, it's user-friendly. My initial features from 6 months ago laid the groundwork for subsequent iterations, now far more sophisticated but proprietary. This journey has been more educational than profitable, appreciating the support from the community. Exploring different strategies continues to be a key focus.

@quant_feed

Understanding the Different Types of PCA: A Simplified Guide

PtrPomorski @ twitter

orig

PCA can often create confusion, but it’s crucial to distinguish between its variants. First, standard PCA is exclusively linear. KernelPCA introduces non-linearity by utilizing a selected kernel, expanding analytical capabilities. For massive datasets, IncrementalPCA serves as a linear PCA alternative that processes data in chunks. Lastly, SparsePCA focuses on feature selection, promoting interpretability by returning a sparse solution. Understanding these distinctions is fundamental for effective data analysis and model selection.

@quant_feed

Exploring the Predictive Power of Private Equity Data on Public Equity Returns

quantseeker @ twitter

orig

Recent findings indicate that private equity deal data from FactSet can be a valuable predictor of public equity returns, encompassing both general market performance and specific sector outcomes. This underscores the interconnectedness of private and public market dynamics, implying that savvy investors can leverage this data for strategic advantage. Regular analysis and insights into quant investing, shared weekly, can help sharpen this investment edge.

@quant_feed

Understanding the Fed Dot Plot: A Critical Look at Economic Projections and Their Accuracy

gametheorizing @ twitter

orig

The Fed's Economic Projections often miss the mark, illustrating that their members struggle to accurately predict their future decisions. This reflects a broader societal issue, where honesty and transparency are compromised for the sake of managing expectations. The Summary of Economic Projections (SEP) isn't genuinely about forecasting; it serves more as a tool to influence markets and shape business sentiment through media engagement.

The jargon used by the Fed—akin to a complex dialect—has become familiar territory for me over the years. I've adeptly interpreted their communications, achieving a flawless track record in predicting their moves in 2022. As I gear up to translate their language again in 2023, I'm maintaining significant positions with substantial CME open interest for upcoming rate decisions. My strategic foresight remains grounded in a deep understanding of the Fed's narrative and intentions.

@quant_feed

Reflecting on My First Day in the FX Market: A Journey Beginning with Harlow Meyer Savage

salr_nyc @ twitter

orig

December 12, 1988 marked my entry into the FX market, kicking off with a successful navigation of the essentials, including finding the toilets and attending the company Christmas party.

Reflecting on that time, I remember the HFT operators actively engaging at the Harlow Meyer Savage USDJPY desk in 2 WTC. It's fascinating how the brokers on that desk often held larger positions than many traders in banks due to the absence of balance sheets, a factor that caught the regulators' attention and led to the push for everyone to operate from screens.

Notably, the talent pool was dynamic; many adept spot brokers transitioned smoothly into FX spot trading roles at banks. It's intriguing to consider the comparative approach of players like Alameda, which didn't seem markedly more strategic to me. In this evolving landscape, I assert that success is in my hands to define.

@quant_feed