Fernando de Meer: Machine Learning to Create Synthetic Financial Time Series
Marriott Hotel Canary Wharf
Wednesday, June 26, 2019 - 06:30 PM
Until: Wednesday, June 26, 2019 - 08:00 PM
(Adjusted for timezone: Europe/London)
ABSTRACT
The scarcity of historical financial data has been a huge hindrance for the development algorithmic trading models ever since the first models were devised. In the ever-changing economic reality we live in, countless models are tried and evaluated. Most of these models seek extracting information from the market by measuring a set of reasonable variables. Through backtesting, an overwhelming amount of these models are seen not to perform and are thus routinely discarded; some however do appear to work well. Out of these models, how many of their performances are just a product of overfitting? Given the lack of available historic data, this is a well-founded concern.
We want to avoid overfitting of investment strategies, and produce a rich enough environment in which to test strategies to find out the most robust.
In today’s world, this is merely wishful thinking, as we have about 55 years of high-quality equity data (or less than 700 monthly observations) for many of the metrics of US stocks we may wish to consider, much less for other types of equity.
In addition to the problem of overfitting, available data is far too small for most machine learning applications and impossibly small for advanced approaches such as deep learning or reinforcement learning. This is particularly troublesome, as Machine Learning approaches do not impose economic principles. If they work, they work in retrospect but not necessarily in the future, since the lack of interpretation and in most cases explicability.
This is where data generation comes into play. Generative Adversarial Networks, GANs, are a type of neural network architecture that have a huge potential in this regard, because they can learn to mimic any distribution of data. That is, GANs can be taught to create worlds pretty similar to our own in any domain: images, music and as we'll see, financial time series.
BIO
Fernando de Meer Pardo is currently a Junior Data Scientist at ETS Asset Management Factory and a MSc student at TU Delft, writing his MSc Thesis under the supervision of C.W.Osterlee on Financial Data Augmentation with GANs. Fernando holds as well a BSc on Mathematics from Madrid Complutense University.
Cost: £15
For more information, visit https://www.meetup.com/thalesians/events/261074078/
The scarcity of historical financial data has been a huge hindrance for the development algorithmic trading models ever since the first models were devised. In the ever-changing economic reality we live in, countless models are tried and evaluated. Most of these models seek extracting information from the market by measuring a set of reasonable variables. Through backtesting, an overwhelming amount of these models are seen not to perform and are thus routinely discarded; some however do appear to work well. Out of these models, how many of their performances are just a product of overfitting? Given the lack of available historic data, this is a well-founded concern.
We want to avoid overfitting of investment strategies, and produce a rich enough environment in which to test strategies to find out the most robust.
In today’s world, this is merely wishful thinking, as we have about 55 years of high-quality equity data (or less than 700 monthly observations) for many of the metrics of US stocks we may wish to consider, much less for other types of equity.
In addition to the problem of overfitting, available data is far too small for most machine learning applications and impossibly small for advanced approaches such as deep learning or reinforcement learning. This is particularly troublesome, as Machine Learning approaches do not impose economic principles. If they work, they work in retrospect but not necessarily in the future, since the lack of interpretation and in most cases explicability.
This is where data generation comes into play. Generative Adversarial Networks, GANs, are a type of neural network architecture that have a huge potential in this regard, because they can learn to mimic any distribution of data. That is, GANs can be taught to create worlds pretty similar to our own in any domain: images, music and as we'll see, financial time series.
BIO
Fernando de Meer Pardo is currently a Junior Data Scientist at ETS Asset Management Factory and a MSc student at TU Delft, writing his MSc Thesis under the supervision of C.W.Osterlee on Financial Data Augmentation with GANs. Fernando holds as well a BSc on Mathematics from Madrid Complutense University.
Cost: £15
For more information, visit https://www.meetup.com/thalesians/events/261074078/