Catch & Signal Release: Hooking Alpha, Releaing Noise
Aug 15, 2025
The Kickoff
August is summer break for most, but if you’re like us, the work hasn’t slowed. Margin–cost spreads in APAC and North America are moving in opposite directions, new datasets are adding less obvious angles, and our conversation with Dr. Paul Bilokon underscored the value of infrastructure that delivers results in live conditions. Welcome to our summer edition, written for quants at their desk or their deck chair.
The Compass
Here's a rundown of what you can find in this edition:
A postcard with our most recent insights
Hot takes from our chat with Dr. Paul Bilokon
APAC Margins Take an Extended Holiday from Costs
If you're fishing for alpha, this dataset might help
All-you-can-read (and listen) buffet
Something chuckle-worthy to add to the holiday mood
Postcard from HQ
It’s been a month of preparation and building momentum. Much of our work has been behind the scenes, setting up the pieces that will come together over the next couple of weeks. On the product side, we’ve expanded our feature library by another 110,000, giving quants more signals to test across diverse market conditions. The UX/UI refresh we began last month is progressing well, now extending to other digital touchpoints to create a more cohesive experience across everything we do and aligning with our longer-term vision.
We’ve also been preparing for our mid-September event in New York with Rebellion and Databento, as a warm up to the CFEM X Rebellion conference on September 19. Alongside that, September will see the launch of new and expanded products, with details to follow soon. And in the background, Quanted Roundups are preparing for a new era, with major changes in store for how we curate and share research with our audience.
The pieces have been coming together all summer, and September will show the results. We're looking forward to sharing more on this soon.
Expert Hot Takes
We recently had the chance to catch up with Dr. Paul Bilokon, someone whose name will be familiar to many in the quant world. He’s the founder and CEO of Thalesians, which he began while still working in industry and has since grown into an international community for collaboration and research at the intersection of AI, quantitative finance, and cybernetics, with a growing community in London, New York, Paris, Frankfurt, and beyond. He is also a visiting professor at Imperial College London and a Research Staff Member at the university’s Centre for Cryptocurrency Research and Engineering, where he focuses on DeFi and blockchain, exploring cryptographic algorithms and inefficiencies in digital asset markets.
Before turning his focus fully to academia and research, Paul spent over a decade on the sell side, building trading systems and leading quant teams across Morgan Stanley, Lehman Brothers, Nomura Citi, and Deutsche Bank, where he played a key role in developing electronic credit trading. Recognised as Quant of the Year in 2023, Paul has built a remarkable career on bridging academic depth with real-world application.
In our conversation, Paul shares how his experience on trading desks shaped his thinking, what excites him about the future of AI in finance, and why practical results still matter most in both research and application.
Having built algorithmic trading systems across FX and credit at institutions like Citi and Deutsche, what stands out to you as the most defining shift in how quant strategies are built and deployed since you entered the field?
I would love to say that there has been a shift towards slick, reliable deployment infrastructures, but this isn’t universally the case: many organisations (I won’t name them) remain pretty bad at infrastructure, making the same mistakes as those mentioned by Fred Brooks in his Mythical Man Month. The successful ones, though, have learned the importance of infrastructure and that it pays to invest in frameworks just as much as it pays to invest in alpha. Such frameworks are well engineered, avoid spurious complexity and hide inevitable complexity, they are easy to extend (including when markets undergo transformative change) and, in the words of my former boss Martin Zinkin, “are easy to use correctly and difficult to use incorrectly.” Another boss of mine (I won’t name him as he likes to keep a low profile) points out the importance of adhering to Uncle Bob’s SOLID principles - many organisations have learned this lesson the hard way, although it’s always preferable to learn from the mistakes of others. Agile techniques are now universally accepted…
What principles or technical habits from your time on trading desks have stayed with you as you moved into research leadership, teaching, and advisory work?
I haven’t really moved anywhere in the sense that I continue to trade, where appropriate to lead, teach, write, and advise. Let me perhaps highlight one of the lessons from trading desks that is particularly useful in all kinds of academic work: it’s knowing what works, what doesn’t work, and where to look for stuff that does work - and keeping things simple and results-oriented. When you own the PnL number, either on your own or jointly, you are naturally motivated by results, rather than by intellectual beauty, etc. So you get stuff done. This is something that was hammered into my head early on, since the days I was a mere Analyst. When you bring this to the academe, while keeping the intellectual rigour, the result is the underrated practically useful research. I’m not necessarily saying that all research is practically useful, but it’s a good feeling when some of your research finds significant applications.
Having worked at the intersection of quant teams, infrastructure, and AI, where do you see the greatest room for improvement in how firms move from research to live deployment?
Statistical rigor and attention to the (usually significant) possibility of overfitting come to mind. People are now acutely aware of the various cognitive biases and logical fallacies that lead to adverse results, and they compensate for that, which is good to see. Your infrastructure should make it difficult to mistranslate what you have done in research when you go to production, so this step should be smooth. Some refer to this as the live-backtest invariance, and some frameworks support it. I do find that putting quants under the pressures of Scrum and Kanban is sometimes productive, sometimes less so. Much of research work is nonlinear and involves leaps of intuition and nonlinear working habits (such as working in bursts or seeking inspiration from a walk in the park). Quants are geniuses, and they should be respected as such. Quite often we try to fit them into a conveyor belt. I would say that the system should indeed have conveyor belts here and there, but it has been a mistake to let go of the legendary Google do-what-you-like Fridays. They aren’t a gimmick, they are genuinely useful in the quant world.
As ML/AI becomes more integrated into quant workflows, what shifts do you expect in how predictive models are designed, interpreted, or monitored?
First and foremost we are talking about more automation, which is often a good thing, except where it’s not. The idea that unskilled people can operate AI-driven systems is an illusion. In object-oriented programming we talk about encapsulation as a means of hiding complexity. But I don’t know a single object-oriented programmer who never had to break encapsulation to check what’s inside. The greatest risk here is the Newtonian vulgar Mechanick: a person who thoughtlessly feeds data into AI models and then uncritically processes the results. This is also known as a Chinese room and considered harmful. I’m an expert precisely because I value my agency, not because I’m given a particular set of tools.
At Thalesians, you’ve supported firms working with high-frequency and complex datasets. What recurring challenges do you see in how teams handle data for research and signal generation?
One of the challenges that we see in this space is siloing. Domain knowledge, technical expertise needed for high-frequency data handling, and mathematical versatility often don’t co-exist, they are often relegated to particular silos. Managers should understand these things intimately and not at the level of buzzwords. Thalesians Ltd. often act as translators, as we speak all these languages.
Anything else you’d like to highlight for those who want to learn more, before we wrap up?
This is a great opportunity. First, I would like to invite the readers to join our MSc in Mathematics and Finance at Imperial College London, to the best of my knowledge the best such programme available anywhere in the world. If you dare of course. This is Imperial College London, and you should be pretty damn good. (If you are reading this, chances are that you are.)
If full-time education is not your thing at this stage of your life, I would like to invite you to the evening courses that Thalesians run with WBS: the Quantitative Developer Certificate and the Machine Learning Certificate in Finance.
There are a few things that I would like to highlight on my SSRN page: particularly the tail-aware approach to resource allocation developed in collaboration with Valeriya Varlashova; what I call deep econometrics; what I call topological number theory. Of course, your feedback is always welcome and actively encouraged.
And on a completely different but personally important note, the archive of Professor Paul Gordon James—long thought lost—has now been unearthed and made available for the first time. Born in Bath, Somerset, on July 4, 1870, Prof. James pursued his education at Christ Church, Oxford, before continuing his postgraduate studies at the Royal College of Science (now part of Imperial College London) under the guidance of Prof. Reginald Dixon.
The collection of his papers, chronicled in The Night’s Cipher, provides remarkable insights into 19th-century history, the nature of consciousness, and artificial intelligence—along with shocking revelations that may finally expose the true identity of Jack the Ripper. Equally controversial are the records concerning Prof. Dixon, a man whose actions remain the subject of fierce debate.
A collective of scholars, recognising the profound historical and philosophical implications of these writings, has taken it upon themselves to preserve and publish them. With the support of Atmosphere Press, Prof. James’s long-hidden work is now available to the public, allowing readers to explore the mysteries he left behind and determine the truth for themselves. You can learn more about this here.
Summer Figures
APAC Margin and Cost Trends at Multi-Year Extremes
July’s electronics manufacturing survey shows a sharp regional profitability spread. The APAC profit-margin diffusion index prints at 125 versus 88 in North America, a 37-point gap. Six-month expectations widen the spread to 34 points (129 versus 95), suggesting the divergence is expected to persist.
The underlying cost structure partly explains the gap. Fourteen percent of APAC firms report falling material costs compared with zero in North America. Labor cost relief is also unique to APAC, with 14 percent expecting declines, again compared with zero in North America. The APAC material cost index stands at 107 now, with forward expectations at 129 — a 22-point rise — indicating expected cost increases rather than declines. North America moves from 134 to 126, an 8-point drop, but from a much higher starting level, leaving net input-cost breadth still materially above APAC in current conditions.
From a modelling perspective, the joint margin–cost picture is notable. In APAC, the positive margin momentum in current conditions is paired with lower current cost breadth than North America, though forward cost expectations in APAC turn higher. North America’s setup shows contracting margins in current breadth terms with elevated cost levels, a combination that in past cycles has correlated with softer earnings trends - though the survey itself does not test that link.
For systematic portfolios, the survey’s orders and shipments data show APAC with a +22 gap in orders and +8 in shipments (future vs current) versus North America’s +10 and +13. Any reference to a “cost–margin composite” or percentile rank, as well as backtest hit-rates for long APAC / short North America configurations, comes from external modelling and is not part of the survey’s published results.
If APAC’s current-cost advantage continues alongside stronger margin breadth, while North America remains cost-pressured in current conditions, the setup could align with sustained cross-regional return differentials into the next reporting cycle - provided forward cost expectations in APAC soften rather than follow the current projected rise.
Source: Global Electronics Association: July 2025 Global Sentiment Report
Data Worth Your Downtime
To some of us, nothing says summer like a fishing trip, but to others the real catch of fishing is the available datasets. Global Fishing Watch offers AIS-based vessel activity, port visits, loitering, encounters, and SAR detections, all delivered through APIs with near real-time refresh. This allows for building signals for seafood supply, identifying illicit transshipment risk, modeling port congestion, and nowcasting coastal macro indicators. They have API packages in both python and R, allowing incorporation into factor models, ESG screens, and macro frameworks with globally consistent, time-stamped coverage. Undoubtedly, It’s a maritime catch worth reeling in for its niche investment potential.
On The Lounger
We know you're probably still thinking about work anyway. Here's some stuff to keep your mind busy when you're supposed to be doing nothing but can't quite turn off the mental models:
📚 FIASCO by Frank Partnoy
📚 Stabilising an Unstable Economy by Hyman Minsky
📚 The Art of Doing Science and Engineering: Learning to Learn by Richard W. Hamming
📚 Where are the Customer's Yachts by Fred Schwed
📚 The Hedge Fund Investing Chartbook: Quantitative Perspectives on the Modern Hedge Fund Investing Experience
📄 Super upside factor by Daniel Shin Un Kang
📄 The behavioural biases of fund managers by Joachim Klement
📄 Probability vs. Likelihood: The Most Misunderstood Duo in Data Science by Unicorn Day
📄 Diving into std::function by Ng Song Guan
📄 The Limits of Out-of-Sample Testing by Nam Nguyen Ph.D.
🎧 Quant Trading: How Hedge Funds Use Data | Marco Aboav, Etna Research by George Aliferis, Investology
🎧 The Psychology of Human Misjudgment by We Study Billionaires
🎧 Laurens Bensdorp - Running 55+ Systematic Trading Strategies Simultaneously by Chat with Traders
🎧 Searching for Signals: BlackRock’s Raffaele Savi on the Future of Systematic Investing by Goldman Sach's Exchanges
🎧 Vinesh Jha: The craft of mining alternative data by The Curious Quant
Finance Fun Corner
——
Disclaimer, this newsletter is for educational purposes only and does not constitute financial advice. Any trading strategy discussed is hypothetical, and past performance is not indicative of future results. Before making any investment decisions, please conduct thorough research and consult with a qualified financial professional. Remember, all investments carry risk
The Kickoff
August is summer break for most, but if you’re like us, the work hasn’t slowed. Margin–cost spreads in APAC and North America are moving in opposite directions, new datasets are adding less obvious angles, and our conversation with Dr. Paul Bilokon underscored the value of infrastructure that delivers results in live conditions. Welcome to our summer edition, written for quants at their desk or their deck chair.
The Compass
Here's a rundown of what you can find in this edition:
A postcard with our most recent insights
Hot takes from our chat with Dr. Paul Bilokon
APAC Margins Take an Extended Holiday from Costs
If you're fishing for alpha, this dataset might help
All-you-can-read (and listen) buffet
Something chuckle-worthy to add to the holiday mood
Postcard from HQ
It’s been a month of preparation and building momentum. Much of our work has been behind the scenes, setting up the pieces that will come together over the next couple of weeks. On the product side, we’ve expanded our feature library by another 110,000, giving quants more signals to test across diverse market conditions. The UX/UI refresh we began last month is progressing well, now extending to other digital touchpoints to create a more cohesive experience across everything we do and aligning with our longer-term vision.
We’ve also been preparing for our mid-September event in New York with Rebellion and Databento, as a warm up to the CFEM X Rebellion conference on September 19. Alongside that, September will see the launch of new and expanded products, with details to follow soon. And in the background, Quanted Roundups are preparing for a new era, with major changes in store for how we curate and share research with our audience.
The pieces have been coming together all summer, and September will show the results. We're looking forward to sharing more on this soon.
Expert Hot Takes
We recently had the chance to catch up with Dr. Paul Bilokon, someone whose name will be familiar to many in the quant world. He’s the founder and CEO of Thalesians, which he began while still working in industry and has since grown into an international community for collaboration and research at the intersection of AI, quantitative finance, and cybernetics, with a growing community in London, New York, Paris, Frankfurt, and beyond. He is also a visiting professor at Imperial College London and a Research Staff Member at the university’s Centre for Cryptocurrency Research and Engineering, where he focuses on DeFi and blockchain, exploring cryptographic algorithms and inefficiencies in digital asset markets.
Before turning his focus fully to academia and research, Paul spent over a decade on the sell side, building trading systems and leading quant teams across Morgan Stanley, Lehman Brothers, Nomura Citi, and Deutsche Bank, where he played a key role in developing electronic credit trading. Recognised as Quant of the Year in 2023, Paul has built a remarkable career on bridging academic depth with real-world application.
In our conversation, Paul shares how his experience on trading desks shaped his thinking, what excites him about the future of AI in finance, and why practical results still matter most in both research and application.
Having built algorithmic trading systems across FX and credit at institutions like Citi and Deutsche, what stands out to you as the most defining shift in how quant strategies are built and deployed since you entered the field?
I would love to say that there has been a shift towards slick, reliable deployment infrastructures, but this isn’t universally the case: many organisations (I won’t name them) remain pretty bad at infrastructure, making the same mistakes as those mentioned by Fred Brooks in his Mythical Man Month. The successful ones, though, have learned the importance of infrastructure and that it pays to invest in frameworks just as much as it pays to invest in alpha. Such frameworks are well engineered, avoid spurious complexity and hide inevitable complexity, they are easy to extend (including when markets undergo transformative change) and, in the words of my former boss Martin Zinkin, “are easy to use correctly and difficult to use incorrectly.” Another boss of mine (I won’t name him as he likes to keep a low profile) points out the importance of adhering to Uncle Bob’s SOLID principles - many organisations have learned this lesson the hard way, although it’s always preferable to learn from the mistakes of others. Agile techniques are now universally accepted…
What principles or technical habits from your time on trading desks have stayed with you as you moved into research leadership, teaching, and advisory work?
I haven’t really moved anywhere in the sense that I continue to trade, where appropriate to lead, teach, write, and advise. Let me perhaps highlight one of the lessons from trading desks that is particularly useful in all kinds of academic work: it’s knowing what works, what doesn’t work, and where to look for stuff that does work - and keeping things simple and results-oriented. When you own the PnL number, either on your own or jointly, you are naturally motivated by results, rather than by intellectual beauty, etc. So you get stuff done. This is something that was hammered into my head early on, since the days I was a mere Analyst. When you bring this to the academe, while keeping the intellectual rigour, the result is the underrated practically useful research. I’m not necessarily saying that all research is practically useful, but it’s a good feeling when some of your research finds significant applications.
Having worked at the intersection of quant teams, infrastructure, and AI, where do you see the greatest room for improvement in how firms move from research to live deployment?
Statistical rigor and attention to the (usually significant) possibility of overfitting come to mind. People are now acutely aware of the various cognitive biases and logical fallacies that lead to adverse results, and they compensate for that, which is good to see. Your infrastructure should make it difficult to mistranslate what you have done in research when you go to production, so this step should be smooth. Some refer to this as the live-backtest invariance, and some frameworks support it. I do find that putting quants under the pressures of Scrum and Kanban is sometimes productive, sometimes less so. Much of research work is nonlinear and involves leaps of intuition and nonlinear working habits (such as working in bursts or seeking inspiration from a walk in the park). Quants are geniuses, and they should be respected as such. Quite often we try to fit them into a conveyor belt. I would say that the system should indeed have conveyor belts here and there, but it has been a mistake to let go of the legendary Google do-what-you-like Fridays. They aren’t a gimmick, they are genuinely useful in the quant world.
As ML/AI becomes more integrated into quant workflows, what shifts do you expect in how predictive models are designed, interpreted, or monitored?
First and foremost we are talking about more automation, which is often a good thing, except where it’s not. The idea that unskilled people can operate AI-driven systems is an illusion. In object-oriented programming we talk about encapsulation as a means of hiding complexity. But I don’t know a single object-oriented programmer who never had to break encapsulation to check what’s inside. The greatest risk here is the Newtonian vulgar Mechanick: a person who thoughtlessly feeds data into AI models and then uncritically processes the results. This is also known as a Chinese room and considered harmful. I’m an expert precisely because I value my agency, not because I’m given a particular set of tools.
At Thalesians, you’ve supported firms working with high-frequency and complex datasets. What recurring challenges do you see in how teams handle data for research and signal generation?
One of the challenges that we see in this space is siloing. Domain knowledge, technical expertise needed for high-frequency data handling, and mathematical versatility often don’t co-exist, they are often relegated to particular silos. Managers should understand these things intimately and not at the level of buzzwords. Thalesians Ltd. often act as translators, as we speak all these languages.
Anything else you’d like to highlight for those who want to learn more, before we wrap up?
This is a great opportunity. First, I would like to invite the readers to join our MSc in Mathematics and Finance at Imperial College London, to the best of my knowledge the best such programme available anywhere in the world. If you dare of course. This is Imperial College London, and you should be pretty damn good. (If you are reading this, chances are that you are.)
If full-time education is not your thing at this stage of your life, I would like to invite you to the evening courses that Thalesians run with WBS: the Quantitative Developer Certificate and the Machine Learning Certificate in Finance.
There are a few things that I would like to highlight on my SSRN page: particularly the tail-aware approach to resource allocation developed in collaboration with Valeriya Varlashova; what I call deep econometrics; what I call topological number theory. Of course, your feedback is always welcome and actively encouraged.
And on a completely different but personally important note, the archive of Professor Paul Gordon James—long thought lost—has now been unearthed and made available for the first time. Born in Bath, Somerset, on July 4, 1870, Prof. James pursued his education at Christ Church, Oxford, before continuing his postgraduate studies at the Royal College of Science (now part of Imperial College London) under the guidance of Prof. Reginald Dixon.
The collection of his papers, chronicled in The Night’s Cipher, provides remarkable insights into 19th-century history, the nature of consciousness, and artificial intelligence—along with shocking revelations that may finally expose the true identity of Jack the Ripper. Equally controversial are the records concerning Prof. Dixon, a man whose actions remain the subject of fierce debate.
A collective of scholars, recognising the profound historical and philosophical implications of these writings, has taken it upon themselves to preserve and publish them. With the support of Atmosphere Press, Prof. James’s long-hidden work is now available to the public, allowing readers to explore the mysteries he left behind and determine the truth for themselves. You can learn more about this here.
Summer Figures
APAC Margin and Cost Trends at Multi-Year Extremes
July’s electronics manufacturing survey shows a sharp regional profitability spread. The APAC profit-margin diffusion index prints at 125 versus 88 in North America, a 37-point gap. Six-month expectations widen the spread to 34 points (129 versus 95), suggesting the divergence is expected to persist.
The underlying cost structure partly explains the gap. Fourteen percent of APAC firms report falling material costs compared with zero in North America. Labor cost relief is also unique to APAC, with 14 percent expecting declines, again compared with zero in North America. The APAC material cost index stands at 107 now, with forward expectations at 129 — a 22-point rise — indicating expected cost increases rather than declines. North America moves from 134 to 126, an 8-point drop, but from a much higher starting level, leaving net input-cost breadth still materially above APAC in current conditions.
From a modelling perspective, the joint margin–cost picture is notable. In APAC, the positive margin momentum in current conditions is paired with lower current cost breadth than North America, though forward cost expectations in APAC turn higher. North America’s setup shows contracting margins in current breadth terms with elevated cost levels, a combination that in past cycles has correlated with softer earnings trends - though the survey itself does not test that link.
For systematic portfolios, the survey’s orders and shipments data show APAC with a +22 gap in orders and +8 in shipments (future vs current) versus North America’s +10 and +13. Any reference to a “cost–margin composite” or percentile rank, as well as backtest hit-rates for long APAC / short North America configurations, comes from external modelling and is not part of the survey’s published results.
If APAC’s current-cost advantage continues alongside stronger margin breadth, while North America remains cost-pressured in current conditions, the setup could align with sustained cross-regional return differentials into the next reporting cycle - provided forward cost expectations in APAC soften rather than follow the current projected rise.
Source: Global Electronics Association: July 2025 Global Sentiment Report
Data Worth Your Downtime
To some of us, nothing says summer like a fishing trip, but to others the real catch of fishing is the available datasets. Global Fishing Watch offers AIS-based vessel activity, port visits, loitering, encounters, and SAR detections, all delivered through APIs with near real-time refresh. This allows for building signals for seafood supply, identifying illicit transshipment risk, modeling port congestion, and nowcasting coastal macro indicators. They have API packages in both python and R, allowing incorporation into factor models, ESG screens, and macro frameworks with globally consistent, time-stamped coverage. Undoubtedly, It’s a maritime catch worth reeling in for its niche investment potential.
On The Lounger
We know you're probably still thinking about work anyway. Here's some stuff to keep your mind busy when you're supposed to be doing nothing but can't quite turn off the mental models:
📚 FIASCO by Frank Partnoy
📚 Stabilising an Unstable Economy by Hyman Minsky
📚 The Art of Doing Science and Engineering: Learning to Learn by Richard W. Hamming
📚 Where are the Customer's Yachts by Fred Schwed
📚 The Hedge Fund Investing Chartbook: Quantitative Perspectives on the Modern Hedge Fund Investing Experience
📄 Super upside factor by Daniel Shin Un Kang
📄 The behavioural biases of fund managers by Joachim Klement
📄 Probability vs. Likelihood: The Most Misunderstood Duo in Data Science by Unicorn Day
📄 Diving into std::function by Ng Song Guan
📄 The Limits of Out-of-Sample Testing by Nam Nguyen Ph.D.
🎧 Quant Trading: How Hedge Funds Use Data | Marco Aboav, Etna Research by George Aliferis, Investology
🎧 The Psychology of Human Misjudgment by We Study Billionaires
🎧 Laurens Bensdorp - Running 55+ Systematic Trading Strategies Simultaneously by Chat with Traders
🎧 Searching for Signals: BlackRock’s Raffaele Savi on the Future of Systematic Investing by Goldman Sach's Exchanges
🎧 Vinesh Jha: The craft of mining alternative data by The Curious Quant
Finance Fun Corner
——
Disclaimer, this newsletter is for educational purposes only and does not constitute financial advice. Any trading strategy discussed is hypothetical, and past performance is not indicative of future results. Before making any investment decisions, please conduct thorough research and consult with a qualified financial professional. Remember, all investments carry risk
The Kickoff
August is summer break for most, but if you’re like us, the work hasn’t slowed. Margin–cost spreads in APAC and North America are moving in opposite directions, new datasets are adding less obvious angles, and our conversation with Dr. Paul Bilokon underscored the value of infrastructure that delivers results in live conditions. Welcome to our summer edition, written for quants at their desk or their deck chair.
The Compass
Here's a rundown of what you can find in this edition:
A postcard with our most recent insights
Hot takes from our chat with Dr. Paul Bilokon
APAC Margins Take an Extended Holiday from Costs
If you're fishing for alpha, this dataset might help
All-you-can-read (and listen) buffet
Something chuckle-worthy to add to the holiday mood
Postcard from HQ
It’s been a month of preparation and building momentum. Much of our work has been behind the scenes, setting up the pieces that will come together over the next couple of weeks. On the product side, we’ve expanded our feature library by another 110,000, giving quants more signals to test across diverse market conditions. The UX/UI refresh we began last month is progressing well, now extending to other digital touchpoints to create a more cohesive experience across everything we do and aligning with our longer-term vision.
We’ve also been preparing for our mid-September event in New York with Rebellion and Databento, as a warm up to the CFEM X Rebellion conference on September 19. Alongside that, September will see the launch of new and expanded products, with details to follow soon. And in the background, Quanted Roundups are preparing for a new era, with major changes in store for how we curate and share research with our audience.
The pieces have been coming together all summer, and September will show the results. We're looking forward to sharing more on this soon.
Expert Hot Takes
We recently had the chance to catch up with Dr. Paul Bilokon, someone whose name will be familiar to many in the quant world. He’s the founder and CEO of Thalesians, which he began while still working in industry and has since grown into an international community for collaboration and research at the intersection of AI, quantitative finance, and cybernetics, with a growing community in London, New York, Paris, Frankfurt, and beyond. He is also a visiting professor at Imperial College London and a Research Staff Member at the university’s Centre for Cryptocurrency Research and Engineering, where he focuses on DeFi and blockchain, exploring cryptographic algorithms and inefficiencies in digital asset markets.
Before turning his focus fully to academia and research, Paul spent over a decade on the sell side, building trading systems and leading quant teams across Morgan Stanley, Lehman Brothers, Nomura Citi, and Deutsche Bank, where he played a key role in developing electronic credit trading. Recognised as Quant of the Year in 2023, Paul has built a remarkable career on bridging academic depth with real-world application.
In our conversation, Paul shares how his experience on trading desks shaped his thinking, what excites him about the future of AI in finance, and why practical results still matter most in both research and application.
Having built algorithmic trading systems across FX and credit at institutions like Citi and Deutsche, what stands out to you as the most defining shift in how quant strategies are built and deployed since you entered the field?
I would love to say that there has been a shift towards slick, reliable deployment infrastructures, but this isn’t universally the case: many organisations (I won’t name them) remain pretty bad at infrastructure, making the same mistakes as those mentioned by Fred Brooks in his Mythical Man Month. The successful ones, though, have learned the importance of infrastructure and that it pays to invest in frameworks just as much as it pays to invest in alpha. Such frameworks are well engineered, avoid spurious complexity and hide inevitable complexity, they are easy to extend (including when markets undergo transformative change) and, in the words of my former boss Martin Zinkin, “are easy to use correctly and difficult to use incorrectly.” Another boss of mine (I won’t name him as he likes to keep a low profile) points out the importance of adhering to Uncle Bob’s SOLID principles - many organisations have learned this lesson the hard way, although it’s always preferable to learn from the mistakes of others. Agile techniques are now universally accepted…
What principles or technical habits from your time on trading desks have stayed with you as you moved into research leadership, teaching, and advisory work?
I haven’t really moved anywhere in the sense that I continue to trade, where appropriate to lead, teach, write, and advise. Let me perhaps highlight one of the lessons from trading desks that is particularly useful in all kinds of academic work: it’s knowing what works, what doesn’t work, and where to look for stuff that does work - and keeping things simple and results-oriented. When you own the PnL number, either on your own or jointly, you are naturally motivated by results, rather than by intellectual beauty, etc. So you get stuff done. This is something that was hammered into my head early on, since the days I was a mere Analyst. When you bring this to the academe, while keeping the intellectual rigour, the result is the underrated practically useful research. I’m not necessarily saying that all research is practically useful, but it’s a good feeling when some of your research finds significant applications.
Having worked at the intersection of quant teams, infrastructure, and AI, where do you see the greatest room for improvement in how firms move from research to live deployment?
Statistical rigor and attention to the (usually significant) possibility of overfitting come to mind. People are now acutely aware of the various cognitive biases and logical fallacies that lead to adverse results, and they compensate for that, which is good to see. Your infrastructure should make it difficult to mistranslate what you have done in research when you go to production, so this step should be smooth. Some refer to this as the live-backtest invariance, and some frameworks support it. I do find that putting quants under the pressures of Scrum and Kanban is sometimes productive, sometimes less so. Much of research work is nonlinear and involves leaps of intuition and nonlinear working habits (such as working in bursts or seeking inspiration from a walk in the park). Quants are geniuses, and they should be respected as such. Quite often we try to fit them into a conveyor belt. I would say that the system should indeed have conveyor belts here and there, but it has been a mistake to let go of the legendary Google do-what-you-like Fridays. They aren’t a gimmick, they are genuinely useful in the quant world.
As ML/AI becomes more integrated into quant workflows, what shifts do you expect in how predictive models are designed, interpreted, or monitored?
First and foremost we are talking about more automation, which is often a good thing, except where it’s not. The idea that unskilled people can operate AI-driven systems is an illusion. In object-oriented programming we talk about encapsulation as a means of hiding complexity. But I don’t know a single object-oriented programmer who never had to break encapsulation to check what’s inside. The greatest risk here is the Newtonian vulgar Mechanick: a person who thoughtlessly feeds data into AI models and then uncritically processes the results. This is also known as a Chinese room and considered harmful. I’m an expert precisely because I value my agency, not because I’m given a particular set of tools.
At Thalesians, you’ve supported firms working with high-frequency and complex datasets. What recurring challenges do you see in how teams handle data for research and signal generation?
One of the challenges that we see in this space is siloing. Domain knowledge, technical expertise needed for high-frequency data handling, and mathematical versatility often don’t co-exist, they are often relegated to particular silos. Managers should understand these things intimately and not at the level of buzzwords. Thalesians Ltd. often act as translators, as we speak all these languages.
Anything else you’d like to highlight for those who want to learn more, before we wrap up?
This is a great opportunity. First, I would like to invite the readers to join our MSc in Mathematics and Finance at Imperial College London, to the best of my knowledge the best such programme available anywhere in the world. If you dare of course. This is Imperial College London, and you should be pretty damn good. (If you are reading this, chances are that you are.)
If full-time education is not your thing at this stage of your life, I would like to invite you to the evening courses that Thalesians run with WBS: the Quantitative Developer Certificate and the Machine Learning Certificate in Finance.
There are a few things that I would like to highlight on my SSRN page: particularly the tail-aware approach to resource allocation developed in collaboration with Valeriya Varlashova; what I call deep econometrics; what I call topological number theory. Of course, your feedback is always welcome and actively encouraged.
And on a completely different but personally important note, the archive of Professor Paul Gordon James—long thought lost—has now been unearthed and made available for the first time. Born in Bath, Somerset, on July 4, 1870, Prof. James pursued his education at Christ Church, Oxford, before continuing his postgraduate studies at the Royal College of Science (now part of Imperial College London) under the guidance of Prof. Reginald Dixon.
The collection of his papers, chronicled in The Night’s Cipher, provides remarkable insights into 19th-century history, the nature of consciousness, and artificial intelligence—along with shocking revelations that may finally expose the true identity of Jack the Ripper. Equally controversial are the records concerning Prof. Dixon, a man whose actions remain the subject of fierce debate.
A collective of scholars, recognising the profound historical and philosophical implications of these writings, has taken it upon themselves to preserve and publish them. With the support of Atmosphere Press, Prof. James’s long-hidden work is now available to the public, allowing readers to explore the mysteries he left behind and determine the truth for themselves. You can learn more about this here.
Summer Figures
APAC Margin and Cost Trends at Multi-Year Extremes
July’s electronics manufacturing survey shows a sharp regional profitability spread. The APAC profit-margin diffusion index prints at 125 versus 88 in North America, a 37-point gap. Six-month expectations widen the spread to 34 points (129 versus 95), suggesting the divergence is expected to persist.
The underlying cost structure partly explains the gap. Fourteen percent of APAC firms report falling material costs compared with zero in North America. Labor cost relief is also unique to APAC, with 14 percent expecting declines, again compared with zero in North America. The APAC material cost index stands at 107 now, with forward expectations at 129 — a 22-point rise — indicating expected cost increases rather than declines. North America moves from 134 to 126, an 8-point drop, but from a much higher starting level, leaving net input-cost breadth still materially above APAC in current conditions.
From a modelling perspective, the joint margin–cost picture is notable. In APAC, the positive margin momentum in current conditions is paired with lower current cost breadth than North America, though forward cost expectations in APAC turn higher. North America’s setup shows contracting margins in current breadth terms with elevated cost levels, a combination that in past cycles has correlated with softer earnings trends - though the survey itself does not test that link.
For systematic portfolios, the survey’s orders and shipments data show APAC with a +22 gap in orders and +8 in shipments (future vs current) versus North America’s +10 and +13. Any reference to a “cost–margin composite” or percentile rank, as well as backtest hit-rates for long APAC / short North America configurations, comes from external modelling and is not part of the survey’s published results.
If APAC’s current-cost advantage continues alongside stronger margin breadth, while North America remains cost-pressured in current conditions, the setup could align with sustained cross-regional return differentials into the next reporting cycle - provided forward cost expectations in APAC soften rather than follow the current projected rise.
Source: Global Electronics Association: July 2025 Global Sentiment Report
Data Worth Your Downtime
To some of us, nothing says summer like a fishing trip, but to others the real catch of fishing is the available datasets. Global Fishing Watch offers AIS-based vessel activity, port visits, loitering, encounters, and SAR detections, all delivered through APIs with near real-time refresh. This allows for building signals for seafood supply, identifying illicit transshipment risk, modeling port congestion, and nowcasting coastal macro indicators. They have API packages in both python and R, allowing incorporation into factor models, ESG screens, and macro frameworks with globally consistent, time-stamped coverage. Undoubtedly, It’s a maritime catch worth reeling in for its niche investment potential.
On The Lounger
We know you're probably still thinking about work anyway. Here's some stuff to keep your mind busy when you're supposed to be doing nothing but can't quite turn off the mental models:
📚 FIASCO by Frank Partnoy
📚 Stabilising an Unstable Economy by Hyman Minsky
📚 The Art of Doing Science and Engineering: Learning to Learn by Richard W. Hamming
📚 Where are the Customer's Yachts by Fred Schwed
📚 The Hedge Fund Investing Chartbook: Quantitative Perspectives on the Modern Hedge Fund Investing Experience
📄 Super upside factor by Daniel Shin Un Kang
📄 The behavioural biases of fund managers by Joachim Klement
📄 Probability vs. Likelihood: The Most Misunderstood Duo in Data Science by Unicorn Day
📄 Diving into std::function by Ng Song Guan
📄 The Limits of Out-of-Sample Testing by Nam Nguyen Ph.D.
🎧 Quant Trading: How Hedge Funds Use Data | Marco Aboav, Etna Research by George Aliferis, Investology
🎧 The Psychology of Human Misjudgment by We Study Billionaires
🎧 Laurens Bensdorp - Running 55+ Systematic Trading Strategies Simultaneously by Chat with Traders
🎧 Searching for Signals: BlackRock’s Raffaele Savi on the Future of Systematic Investing by Goldman Sach's Exchanges
🎧 Vinesh Jha: The craft of mining alternative data by The Curious Quant
Finance Fun Corner
——
Disclaimer, this newsletter is for educational purposes only and does not constitute financial advice. Any trading strategy discussed is hypothetical, and past performance is not indicative of future results. Before making any investment decisions, please conduct thorough research and consult with a qualified financial professional. Remember, all investments carry risk
