Twenty years of tick data. Fifty seconds.
That was the calculation we shared at STAC, illustrating how quickly historical market data could theoretically be accessed with high-throughput storage.
For quant teams working with years of tick data, retrieval performance can shape how quickly researchers move from historical data to testing a trading hypothesis.
But accessing historical data is only part of the challenge. The real question is what happens after researchers access the data they need.
Why does quant research speed matter before the trade?
In quantitative trading, speed is often associated with execution. Firms invest heavily in shaving picoseconds from transactions, optimizing algorithms, and reducing latency between exchanges. Yet much of the work that determines a trading strategy’s success happens long before an order reaches the market.
Quant researchers develop hypotheses, analyze historical market behavior, test models, and refine strategies. Each experiment requires access to data, often spanning years of trading activity.
The process is iterative. Develop a hypothesis, test it against historical data, examine the results, refine the model, and repeat. The faster that cycle moves, the more opportunities researchers have to explore and validate potential trading strategies.
When data access becomes a bottleneck, the entire process slows. And in a business where trading opportunities can be fleeting, those delays carry consequences beyond researcher productivity.
What slows quant research when infrastructure gets in the way?
Market data rarely moves directly from an exchange feed into a research model. Firms rely on interconnected systems, including Kafka, kdb+, Spark, and other analytics platforms, to ingest, process, store, and analyze information. Each serves a purpose. But as data volumes grow, moving information between systems can introduce delays and operational complexity.
Researchers may spend valuable time waiting for data sets, troubleshooting performance bottlenecks, or resolving differences between research and production environments. These problems become especially apparent when firms move from proof of concept to production.
A proof of concept may perform well with a limited data set and controlled environment. Production, however, introduces larger data sets, concurrent workloads, continuous ingestion, and more demanding performance requirements. Suddenly, researchers who should be developing strategies are managing infrastructure problems instead.
The friction often appears in two places: retrieving the data needed for analysis and maintaining consistency between research and production environments.
Neither generates alpha. Both consume time that could otherwise be spent testing ideas.
What does slow data cost a quant research team?
Consider a quant researcher running a series of backtests. If each test takes hours rather than minutes, fewer hypotheses can be evaluated in a given day.
That limitation changes which ideas receive attention.
Researchers may prioritize strategies they already believe are promising rather than spend time testing more speculative approaches. But some of the most interesting opportunities may come from ideas that initially appear less certain.
When infrastructure limits experimentation, those ideas may never receive a proper evaluation. Faster iteration only helps when it’s paired with discipline. More hypotheses mean more risk of overfitting, so out-of-sample testing and walk-forward validation matter even more as research speeds up.
There is also the question of historical depth.
To reduce processing time, teams may shorten their lookback windows or move older data sets to less expensive storage tiers. A model trained against a narrower historical window may miss market conditions, volatility patterns, or behaviors present in the broader data set.
Then there is timing.
A trading strategy’s edge can decay as market conditions change or competitors discover similar opportunities. A signal that looks promising today may become less valuable by the time it’s validated and deployed. If slow infrastructure extends the research cycle, a firm may have less time to capitalize on that opportunity.
The cost isn’t simply lost hours. It’s the potential alpha that never makes it into production.
For firms employing highly skilled quantitative researchers, that represents two forms of lost value: expensive research capacity spent waiting rather than experimenting, and potential trading opportunities that may never be realized.
How does 23/5 trading change quant research windows?
Beginning December 6, 2026, at 9pm ET, Nasdaq plans to expand US equities trading to 23 hours a day, five days a week. Under the new schedule, trading will pause for just one hour each weekday, from 8pm to 9pm ET.
Historically, firms have relied on overnight windows for data ingestion, reconciliation, reference data updates, backtesting, and model retraining.
As trading hours expand, those windows become smaller, placing greater demands on infrastructure. Quant teams may need to run historical analyses and computationally intensive backtests while production systems continue ingesting and processing live market data.
Firms now have to support research, maintain production performance, and upgrade infrastructure without the maintenance windows they used to rely on.
Firms already designed for continuous operations may be better positioned to adapt. Those dependent on overnight batch processing could face greater challenges. The shift to 23/5 reinforces a broader reality: Infrastructure must support research and production simultaneously, not force teams to choose between them.
What role should storage play in quant research?
Storage performance is traditionally measured through capacity, throughput, latency, availability, and cost. But for quant teams, what matters is whether infrastructure lets researchers access the historical data they need, test more hypotheses, and move promising strategies toward production without unnecessary delays.
This is where Everpure™ FlashBlade//EXA™ enters the conversation.
Designed for demanding AI and high-performance computing workloads, FlashBlade//EXA provides a scale-out storage architecture focused on high-throughput data access.
Research and backtesting may need to run while production systems continue ingesting live market data. A shared scale-out platform is intended to support both at the same time, rather than pushing research into a shrinking overnight window.
Tick data archives are often copied across research, production, and analytics environments. Each copy adds cost and another place for the environments to drift apart. Consolidating onto a single platform may reduce the number of copies a firm has to maintain. Running research and production on one platform may also narrow the gap between the data a model is tested against and the data it runs against.
At STAC, we illustrated its potential using approximately 500TB of historical US equities and options tick data. At an assumed throughput of 10TB/s, that volume could theoretically be read in 50 seconds.
But the bigger question is what quant teams could accomplish if retrieving historical market data were no longer a significant constraint.
Every research team has a list of ideas that never got tested because there was not time to run them properly. Some of those ideas were probably not worth pursuing, but a few may have been the edge a competitor found first. With that constraint gone, the list of untested ideas gets shorter, and the next source of alpha may be one that would otherwise have stayed on it.
Frequently Asked Questions
- How does slow historical data access limit quant research? Each experiment needs data that often spans years, so if backtests take hours instead of minutes, fewer hypotheses get tested, teams may shorten lookback windows or move older data to cheaper tiers, and a strategy’s edge may decay while the research waits. The cost is potential alpha that never reaches production, not only lost hours.
- How does Nasdaq 23/5 trading change overnight processing windows for quant teams? Beginning December 6, 2026 at 9pm ET, Nasdaq plans to trade US equities 23 hours a day, five days a week, pausing only from 8pm to 9pm ET. Windows firms used for ingestion, reconciliation, backtesting and retraining shrink, so research may need to run while production keeps ingesting live data. [SME-pending: source for the historical reliance on overnight windows]
- What storage lets quant research and production run on one platform? Everpure FlashBlade//EXA provides a scale-out architecture focused on high-throughput data access, and a shared platform is intended to support research and production at the same time.
- How fast could 500TB of tick data be read? At an assumed 10TB/s, about 500TB of historical US equities and options tick data could theoretically be read in 50 seconds. This is an illustrative calculation, not a measured result. [SME-pending: origin of the 500TB figure]
- Can consolidating tick data onto one platform reduce copies and drift between research and production? It may. Tick data archives are often copied across research, production and analytics environments, each copy adds cost, and one platform may narrow the gap between the data a model is tested on and the data it runs on.












