Author: Selim Adyel (@selim_adyel)

In order to design the most comprehensive and responsive system for Concrete, we leveraged scientific, rigorous, and robust quantitative techniques.

The Quantitative Framework, built from the ground up for Concrete, harnesses AI and state of the art institutional quantitative techniques and technologies to forecast digital assets prices at different horizons and it aims primarily at pricing probabilities of downside moves.

The foundation of the Concrete Quantitative Framework has been built off of over a decade of R&D in both TradFi and digital assets. Applications of the framework have generated a multi-years outstanding trading track record.

The task of modeling financial time series is notoriously difficult and requires the generalist skillset of a data scientist, developer, researcher, data engineer, and a trader.

Financial data has a low signal to noise ratio, it is patchy and features so called ‘stylized facts’ (more on that in the ‘Opportunities’ section).

In general, most traders and strategies (especially in digital assets) lose money on average or fail to beat a passive benchmark.

From a project management perspective, most common mistakes include: falling for the latest flashy technique without getting the basics right (e.g.: data pipeline), over-engineering, overfitting and scalability issues due to faulty initial design.

Quantitative research is an Intellectual Property heavy discipline where the decisions on selecting and retaining the right human capital can make or break an entire project or company.

Another mistake would be to lose too much time on research papers. Tenure-seeking researchers publish thousands of academic articles that promote dubious investment strategies, without controlling for multiple testing.

“Predicting financial outcomes is orders-of-magnitude more difficult than, for example, recognizing faces or driving cars. When applied incorrectly, ML compounds the risk of backtest overfitting.” - Marco Lopez de Prado

A summary of the most common challenges includes:

1. Wrong expectations

2. Human capital management

3. Poor project management with wrong priorities

4. Lack of investment in the right tools (data, etc.)

5. Neglecting the basics: data pipeline, robust servers

6. Over-engineering and underperforming infrastructure

7. Falling for the latest flashy AI tool without mastering the basics and losing focus

8. Dealing with noisy and patchy data

9. Financial time series specific microstructure and dynamics (stylized facts)

In the particular case of predicting short-term digital assets price predictions, the following additional challenges apply:

• Volatility: digital assets markets are highly volatile in the short term, influenced by a myriad of factors including news, market sentiment, regulatory changes, and technical developments

• Noise: short-term price movements can often be driven by noise and speculative trading rather than fundamental factors, making short-term predictions more susceptible to inaccuracies

There is currently a lack of institutional quality quantitative frameworks in digital assets and more specifically - DeFi - which gives us a competitive edge.

2020 to mid 2021 was a period marked by a large proportion of retail investors in the digital assets (but also equities) market, supported by various helicopter money policies, especially in the USA, but also by savings accumulated during the Covid lockdowns.

Retail traders are emotional and have cognitive bias that affect their trading, such as:

1. Confirmation bias: look for convenient information to support your beliefs or conclusions

2. Hindsight bias: tendency to believe that your past positive outcomes were a result of your ability to understand and predict what the market will do

3. Recency bias: the belief that the current information can influence the future more than the older information, which is generally considered useless

4. Herd mentality

These, in turns, produce what we call in financial engineering ‘stylized facts’, such as:

1. Heavy tails: high likelihood of outlier returns

2. Leverage effect: volatility is negatively correlated to price returns

3. Absence of linear autocorrelation: past returns not correlated with current returns

4. Volatility clustering: large/small changes tend to be followed by larger/small changes respectively

5. Gain/loss asymmetry: financial time series take longer for going up than going down

6. Aggregational Gaussianity: shape of distribution not the same at different time scales

Technical breakout factors are a good way of exploiting market inefficiencies created by these biases.

However, they are too simplistic and not sufficient. More sophisticated tools are required.

The sector has seen an increase of institutional investors and traders. Institutional and experienced traders are more often macro driven, either because they trade other asset classes or because they understand the causal relationships between fiscal policies, flow and macro assets with digital assets price action.

Therefore, the digital assets market is also macro driven.

Statistical arbitrage models are an excellent way of exploiting all these inefficiencies and dynamics.

Statistical arbitrage is a family of forecasting models and trading strategies, leveraging statistical learning and AI. The goal is to, on average, realize more gains than losses, hence having positive expected gains or forecasts.

The Quantitative Framework is made up of a robust foundation of building blocks, allowing for new data sets additions and machine learning models to be used and new models to be developed at different forecast horizons.

The Quantitative Framework aims at being robust and minimizing the risk of false discovery, hence producing models that outperform out of sample.

A pragmatic approach is used, prioritizing empirical evidence, common sense and battle tested methodologies.

We do not build a forecasting model around a machine learning technique (which is a common mistake) but prefer to build a framework based on a robust R&D process; the process is more important than any ML model, otherwise many more people would be good traders.

Datasets currently in production include digital assets price (and its derivatives), global macro futures contracts and ETFs prices feeds across commodities, equities, fixed income and currencies.

We have built an institutional level robust quant database and data pipeline, specifically designed to handle big data and financial time series.

After relevant derived data preparation by leveraging proprietary techniques, datasets are fed through the data pipeline to produce inputs for our models.

We leverage statistical modeling and AI to extract alpha, together with proprietary quantitative techniques.

We currently run seven models into production, with forecasting horizons ranging from hours to weeks, covering the full spectrum of short, medium and long-term horizons.

Alpha is a term used in investing and trading to describe a model (or trading strategy) ability to outperform, or its ‘edge.’

By specifically taking into account live and historical markets prices action and reacting dynamically, Concrete leverages the Quantitative Framework and its proprietary IP to supercharge the protocol.

In particular, it provides Concrete with a competitive edge and makes it more valuable by leveraging state of the art institutional and battle tested quantitative techniques and technologies powered by AI.

If you are interested in chatting about protocol design, working together, or integrating within the Concrete ecosystem, please reach out to hello@concrete.xyz

Follow along on Twitter and join the community.

Concrete Protocol: https://twitter.com/ConcreteXYZ

Blueprint Finance: https://twitter.com/Blueprint_DeFi

Author: Selim Adyel (@selim_adyel)

In order to design the most comprehensive and responsive system for Concrete, we leveraged scientific, rigorous, and robust quantitative techniques.

The Quantitative Framework, built from the ground up for Concrete, harnesses AI and state of the art institutional quantitative techniques and technologies to forecast digital assets prices at different horizons and it aims primarily at pricing probabilities of downside moves.

The foundation of the Concrete Quantitative Framework has been built off of over a decade of R&D in both TradFi and digital assets. Applications of the framework have generated a multi-years outstanding trading track record.

The task of modeling financial time series is notoriously difficult and requires the generalist skillset of a data scientist, developer, researcher, data engineer, and a trader.

Financial data has a low signal to noise ratio, it is patchy and features so called ‘stylized facts’ (more on that in the ‘Opportunities’ section).

In general, most traders and strategies (especially in digital assets) lose money on average or fail to beat a passive benchmark.

From a project management perspective, most common mistakes include: falling for the latest flashy technique without getting the basics right (e.g.: data pipeline), over-engineering, overfitting and scalability issues due to faulty initial design.

Quantitative research is an Intellectual Property heavy discipline where the decisions on selecting and retaining the right human capital can make or break an entire project or company.

Another mistake would be to lose too much time on research papers. Tenure-seeking researchers publish thousands of academic articles that promote dubious investment strategies, without controlling for multiple testing.

“Predicting financial outcomes is orders-of-magnitude more difficult than, for example, recognizing faces or driving cars. When applied incorrectly, ML compounds the risk of backtest overfitting.” - Marco Lopez de Prado

A summary of the most common challenges includes:

1. Wrong expectations

2. Human capital management

3. Poor project management with wrong priorities

4. Lack of investment in the right tools (data, etc.)

5. Neglecting the basics: data pipeline, robust servers

6. Over-engineering and underperforming infrastructure

7. Falling for the latest flashy AI tool without mastering the basics and losing focus

8. Dealing with noisy and patchy data

9. Financial time series specific microstructure and dynamics (stylized facts)

In the particular case of predicting short-term digital assets price predictions, the following additional challenges apply:

• Volatility: digital assets markets are highly volatile in the short term, influenced by a myriad of factors including news, market sentiment, regulatory changes, and technical developments

• Noise: short-term price movements can often be driven by noise and speculative trading rather than fundamental factors, making short-term predictions more susceptible to inaccuracies

There is currently a lack of institutional quality quantitative frameworks in digital assets and more specifically - DeFi - which gives us a competitive edge.

2020 to mid 2021 was a period marked by a large proportion of retail investors in the digital assets (but also equities) market, supported by various helicopter money policies, especially in the USA, but also by savings accumulated during the Covid lockdowns.

Retail traders are emotional and have cognitive bias that affect their trading, such as:

1. Confirmation bias: look for convenient information to support your beliefs or conclusions

2. Hindsight bias: tendency to believe that your past positive outcomes were a result of your ability to understand and predict what the market will do

3. Recency bias: the belief that the current information can influence the future more than the older information, which is generally considered useless

4. Herd mentality

These, in turns, produce what we call in financial engineering ‘stylized facts’, such as:

1. Heavy tails: high likelihood of outlier returns

2. Leverage effect: volatility is negatively correlated to price returns

3. Absence of linear autocorrelation: past returns not correlated with current returns

4. Volatility clustering: large/small changes tend to be followed by larger/small changes respectively

5. Gain/loss asymmetry: financial time series take longer for going up than going down

6. Aggregational Gaussianity: shape of distribution not the same at different time scales

Technical breakout factors are a good way of exploiting market inefficiencies created by these biases.

However, they are too simplistic and not sufficient. More sophisticated tools are required.

The sector has seen an increase of institutional investors and traders. Institutional and experienced traders are more often macro driven, either because they trade other asset classes or because they understand the causal relationships between fiscal policies, flow and macro assets with digital assets price action.

Therefore, the digital assets market is also macro driven.

Statistical arbitrage models are an excellent way of exploiting all these inefficiencies and dynamics.

Statistical arbitrage is a family of forecasting models and trading strategies, leveraging statistical learning and AI. The goal is to, on average, realize more gains than losses, hence having positive expected gains or forecasts.

The Quantitative Framework is made up of a robust foundation of building blocks, allowing for new data sets additions and machine learning models to be used and new models to be developed at different forecast horizons.

The Quantitative Framework aims at being robust and minimizing the risk of false discovery, hence producing models that outperform out of sample.

A pragmatic approach is used, prioritizing empirical evidence, common sense and battle tested methodologies.

We do not build a forecasting model around a machine learning technique (which is a common mistake) but prefer to build a framework based on a robust R&D process; the process is more important than any ML model, otherwise many more people would be good traders.

Datasets currently in production include digital assets price (and its derivatives), global macro futures contracts and ETFs prices feeds across commodities, equities, fixed income and currencies.

We have built an institutional level robust quant database and data pipeline, specifically designed to handle big data and financial time series.

After relevant derived data preparation by leveraging proprietary techniques, datasets are fed through the data pipeline to produce inputs for our models.

We leverage statistical modeling and AI to extract alpha, together with proprietary quantitative techniques.

We currently run seven models into production, with forecasting horizons ranging from hours to weeks, covering the full spectrum of short, medium and long-term horizons.

Alpha is a term used in investing and trading to describe a model (or trading strategy) ability to outperform, or its ‘edge.’

By specifically taking into account live and historical markets prices action and reacting dynamically, Concrete leverages the Quantitative Framework and its proprietary IP to supercharge the protocol.

In particular, it provides Concrete with a competitive edge and makes it more valuable by leveraging state of the art institutional and battle tested quantitative techniques and technologies powered by AI.

If you are interested in chatting about protocol design, working together, or integrating within the Concrete ecosystem, please reach out to hello@concrete.xyz

Follow along on Twitter and join the community.

Concrete Protocol: https://twitter.com/ConcreteXYZ

Blueprint Finance: https://twitter.com/Blueprint_DeFi