I am writing this post as a follow up on a talk by the same name given at Re-work Deep Learning Summit, Singapore. In the talk I tried to detail the reasons why the financial models fail and how deep learning can bridge the gap. Further on, I moved on to present three use cases for deep learning in Finance and evidence of the superiority of these models.

While finance is the most computationally intensive field that there is, the widely used models in finance — the supervised and unsupervised models, the state based models, the econometric models or even the stochastic models are marred by the problems of over fitting, heuristics and poor out of sample results.

Which is because, the financial domain is hugely complex and non-linear with a plethora of factors influencing each other. To solve this, if we look at the research done in Deep Learning in proven fields of image recognition, speech recognition or sentiment analysis we see that these models are capable of learning from large scaled unlabelled data, forming non-linear relationships, forming recurrent structures and can be easily tweaked to avoid over-fitting.

If these models find application in the discipline of finance then the applications are far and wide. These models can be used in pricing, portfolio construction, risk management and even high frequency trading to name a few fields. So let us tackle a few of these problems. Taking the sample problem of predicting daily Gold Prices, we first look at the traditional methods. Using the Autoregressive Integrated moving Average model, which tries to predict a stationary time series keeping the seasonal component in place we get a result.

If we add related predictor variables to our auto-regressive model and move to a Vector Auto Regressive model, we get these results —.

**Tutorial: Deep Reinforcement Learning For Algorithmic Trading in Python**

Deep Regression. Using the same inputs if I fit a simple deep regression model on the data, I get far better results. Convolutional Neural Networks. Modifying my architecture to use convolutional neural networks for the same problem, my results are. These results are drastically better. But the best results come next. There you go! Using these variations of recurrent neural networks, my results are:. So overall the trend of the mean squared errors is a revelation!

The second financial problem we will try to tackle using deep learning is of portfolio construction. The application of deep learning to this problem has a beautiful construct. My study is inspired by a paper titled Deep Portfolios. What the authors of the paper try to do is to construct auto-encoders that map a time series to itself.Artificial Intelligence in Finance Spring Course Information.

This course explores the basic concepts and underlying principles of artificial intelligence AIdelving into the fundamentals of machine learning with insights from case studies of relevant technologies. Allowing for the experimentation of applications of machine learning, this course is designed to encourage students to devise creative ways to put readily-available AI technologies to use to tackle problems in real life.

With a focus on the conceptual understanding of the fundamentals of AI, the purpose of this course is two-fold: 1. To educate the next generation of innovators of the digital economy, especially in the financial sector; 2.

## Deep Learning in Finance

To equip FinTech leaders with the conceptual frameworks and analytical tools for the development and deployment of AI-enabled applications and technologies. Prerequisite : Some preliminary course on statistical machine learning, applied statistics, and deep learning will be helpful. Anthony Woo [ Resume ] Yuan Yao. The system is highly catered to getting you help fast and efficiently from classmates and myself. Rather than emailing questions to the teaching staff, I encourage you to post your questions on Piazza.

If you have any problems or feedback for the developers, email team piazza. By James, Witten, Hastie, and Tibshirani. Python-Numpy Tutorials by Justin Johnson. Jupyter Notebook Tutorials. PyTorch Tutorials. Tensorflow Tutorials. MXNet Tutorials. Theano Tutorials. By Hastie, Tibshirani, and Friedman. Email: Mr. Yifei Huang aifin. Katrina Fong. Lecture History and Overview of Artificial Intelligence. Lecture Chen NING.As a reminder, the purpose of this series of articles is to experiment with state-of-the-art deep reinforcement learning technologies to see if we can create profitable Bitcoin trading bots.

However, recent advances in the field have shown that RL agents are often capable of learning much more than supervised learning agents within the same problem domain. For this reason, I am writing these articles to see just how profitable we can make these trading agents, or if the status quo exists for a reason. Finally, we will use a technique called Bayesian optimization to zone in on the most profitable hyper-parameters, before training and testing the final agents profitablity.

Hold on to your seats everyone, this is going to be a wild ride. The first thing we need to do to improve the profitability of our model, is make a couple improvements on the code we wrote in the last article. If you do not yet have the code, you can grab it from my GitHub.

Instead, it is inherently captured by the recursive nature of the network. At each time step, the input from the data set is passed into the algorithm, along with the output from the last time step.

It was also pointed out to me on the last article that our time series data is not stationaryand therefore, any machine learning model is going to have a hard time predicting future values. A stationary time series is one whose mean, variance, and auto-correlation lagged correlation with itself are constant.

The bottom line is that our time series contains an obvious trend and seasonality, which both impact our algorithms ability to predict the time series accurately. We can fix this by using differencing and transformation techniques to produce a more normal distribution from our existing time series. Differencing is the process of subtracting the derivative rate of return at each time step from the value at that time step.

This has the desired result of removing the trend in our case, however, the data still has a clear seasonality to it. We can attempt to remove that by taking the logarithm at each time step before differencing, which produces the final, stationary time series, shown below on the right.

We can verify the produced time series is stationary by running it through an Augmented Dickey-Fuller Test. Doing this gives us a p-value of 0. To further improve our model, we are going to be doing a bit of feature engineering.

Feature engineering is the process of using domain-specific knowledge to create additional input data that improves a machine learning model. In our case, we are going to be adding some common, yet insightful technical indicators to our data set, as well as the output from the StatsModels SARIMAX prediction model. The technical indicators should add some relevant, though lagging information to our data set, which will be complimented well by the forecasted data from our prediction model. This combination of features should provide a nice balance of useful observations for our model to learn from.

To choose our set of technical indicators, we are going to compare the correlation of all 32 indicators 58 features available in the ta library. We can use pandas to find the correlation between each indicator of the same type momentum, volume, trend, volatilitythen select only the least correlated indicators from each type to use as features.

That way, we can get as much benefit out of these technical indicators as possible, without adding too much noise to our observation space. It turns out that the volatility indicators are all highly correlated, as well as a couple of the momentum indicators.

Next we need to add our prediction model. For example, our agent can be learn to be more cautious trusting predictions when the confidence interval is small and take more risk when the interval is large. One might think our reward function from the previous article i. While our simple reward function from last time was able to profit, it produced volatile strategies that often lead to stark losses in capital. To improve on this, we are going to need to consider other metrics to reward, besides simply unrealized profit.

While this strategy is great at rewarding increased returns, it fails to take into account the risk of producing those high returns.The lecture has been developed by Christa Cuchiero and Josef Teichmann. Several further issues and ramifications are planned. More material will follow soon. The lecture introduces several fundamental concepts from machine learning with a view towards important financial applications.

Lecture notes are provided as ipython notebooks or in form of slides as well as of classical notes. Most of the following code runs savely under Python 3. You can get to a downloadable.

The course in Vienna held by Christa Cuchiero splits into partially distinct parts for Master and PhD students, the structure is similar but more exercises and some slides are included:. References and several talks on new developments can be found at the end of this webpage. Basic material and some exercises for the courses in Zurich in spring and autumn The lecture introduces several fundamental concepts from machine learning with a view towards important financial applications.

Lecture 1 Introduction, Universal Approximation by shallow networks, one string of arguments for depth : Lecture 1 as iPython notebook and some training data should be unpacked, then store the files in the folder of the notebook. For the exercises see also Function approximation with linear models and neural network from Tirthajyoti Sarkar's github resources. Lecture 2 neural ordinary differential equations, backpropagation, expressiveness by randomness : Lecture 2 as iPython notebook and some data-file.

For the exercises just write a two-hidden layer network by hand with backpropagation by hand following the code in the notebook, see also one possible solution. Lecture 3 Deep Hedging without transaction costs : Lecture 3 as iPython notebookwhere a tensorflow implementation of Deep Hedging as well as some background is explained.

A short Keras implementation of deep portfolio optimization without transaction costs, but easily to be modified can be found at as iPython notebook. A Keras implementation of Deep Hedging for the BS model with analysis of the hedging strategies can be found at as iPython notebook. An implementation illustrating the use of Keras in a very simple case can be found here.

Some general thoughts on training can be found here as iPython notebook. A short Keras implementation of deep portfolio optimization with transaction costs you can be found here as iPython notebook.

A Keras implementation of Deep portfolio optimization in the Merton model with analysis of the trading strategies can be found here as iPython notebook. Lecture 5 Deep Simulation : Lecture 5 as iPython notebook.We construct realistic equity option market simulators based on generative adversarial networks GANs.

We consider recurrent and temporal convolutional architectures, and assess the impact of state compression. Option market simulators are highly relevant because they allow us to extend the limited real-world data sets available for the training and evaluation of option trading strategies. We show that network-based generators outperform classical methods on a range of benchmark metrics, and adversarial training achieves the best performance.

Our work demonstrates for the first time that GANs can be successfully applied to the task of generating multivariate financial time series. Magnus Wiese. Lianjun Bai.

Ben Wood. Hans Buehler. Access to medical data is highly restricted due to its sensitive nature, Systematic trading strategies are algorithmic procedures that allocate a Modeling financial time series by stochastic processes is a challenging In this work we introduce QuantNet: an architecture that is capable of t Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

There is growing interest in applying reinforcement learning techniques to the problem of managing a portfolio of derivatives. Unfortunately, the amount of useful real-life data available is limited; if we take a sampling interval of one day, ten years of option prices translates into only a few thousand samples. This motivates the need for a realistic simulator: we can generate much larger volumes of data for training and evaluation while preserving the key distributional properties of the real samples, thus helping to avoid overfitting.

In this article, we build and test a collection of simulators based on neural networks NNs. We begin by transforming the option prices to an equivalent representation in the form of. There is a wide range of existing literature on the generation of synthetic time series for asset prices see, e.

Classical derivative pricing models e. Recently, generative adversarial networks GANs Goodfellow et al. Zhang et al.Over the past few years, deep neural networks have become extremely popular.

This emerging field of computer science was created around the concept of biological neural networks, and deep learning has become something of a buzzword today. Deep learning scientists and engineers try to mathematically describe various patterns from biological nervous systems. Deep learning systems have been applied to various problems: computer vision, speech recognition, natural language processing, machine translation, and more.

It is interesting and exciting that in some tasks, deep learning has outperformed human experts.

### Introduction to Deep Learning Trading in Hedge Funds

Today, we will be taking a look at deep learning in the financial sector. One of the more attractive applications of deep learning is in hedge funds. Hedge funds are investment funds, financial organizations that raise funds from investors and manage them.

They usually work with time series data and try to make some predictions. There is a special type of deep learning architecture that is suitable for time series analysis: recurrent neural networks RNNsor even more specifically, a special type of recurrent neural network: long short-term memory LSTM networks.

LSTMs are capable of capturing the most important features from time series data and modeling its dependencies. A stock price prediction model is presented as an illustrative case study on how hedge funds can use such systems.

PyTorch framework, written in Python, is used to train the model, design experiments, and draw the results. We will start with some deep learning basics before moving on to real-world examples:. One of the most challenging and exciting tasks in the financial industry is predicting whether stock prices will go up or down in the future. Today, we are aware that deep learning algorithms are very good at solving complex tasks, so it is worth trying to experiment with deep learning systems to see whether they can successfully solve the problem of predicting future prices.

Nvidia helped revolutionize deep learning networks a decade ago, as it started offering very fast graphics processing units GPUs for general purpose computing in Tesla-series products. There are very few scientific papers about using deep learning in finance, but demand for deep learning experts from fintech companies is strong, as they obviously recognize its potential.

This article will help explain why deep learning in finance is becoming increasingly popular by outlining how financial data is used in constructing deep learning systems. A special type of recurrent neural network—the LSTM network —will be presented as well.

We will outline how a finance-related task can be solved using recurrent neural networks. This article also features an illustrative case study on how hedge funds can use such systems, presented through experiments. We will also consider how deep learning systems can be improved and how hedge funds can go about hiring talent to build those systems, i. Before we proceed to the technical aspect of the problem, we need to explain what makes hedge funds unique.

So, what is a hedge fund? A hedge fund is an investment fund—a financial organization which raises funds from investors and places them in short-term and long-term investments, or in different financial products. It is typically formed as a limited partnership or a limited liability company. It is generally accepted that when more risk is taken, there is greater potential for higher returns and losses.

In order to achieve good returns, hedge funds rely on various types of investment strategies, trying to make money by exploiting market inefficiencies.

Due to various kinds of investment strategies that are not allowed in ordinary investment funds, hedge funds are not registered as funds, i. Some hedge funds generate more money than the market average, but some of them lose money.

Not just anyone can invest in hedge funds, though. Hedge funds are intended for a small number of wealthy investors. Usually, the ones who want to take part in hedge funds need to be accredited. That means that they must have a special status regarding financial regulation laws.GitHub is home to over 40 million developers working together to host and review code, manage projects, and build software together.

If nothing happens, download GitHub Desktop and try again. If nothing happens, download Xcode and try again. If nothing happens, download the GitHub extension for Visual Studio and try again. One should avoid solving more difficult intermediate problems when solving a target problem. Vladimir Vapnik, Statistical Learning Theory, Stochastic optimal control assumes a model, simulations assume some knowledge of the world say monte carloalternative and more robust simulation methods?

We are looking to allocate to assets or strategies in a manner that is better than the current state of the art and to get RL working in real world finance. Reinforcement learning is a method for solving MDP's in a model free fashion. There are many MDP problems in finance and a whole mathematical methodology such as stochastic optimal control. Solutions may have particular value when there is path dependency on an agent's decisions into the future.

In the derivative hedging method of finance, problems are usually solved in a step-wise fashion Recent paper's hint at the ability to directly learn a hedging strategy in a greek free fashion from a simulation of the environment. In other words rather than a 2 step process - model the environment, solve the model, we can go straight from simulation to hedging, including where there are difficult real life problems such as transaction costs and path dependency and indeed complex risk adjusted functions of our final distribution of returns that we wish to maximise.

Allocation decisions within Finance lie within a most difficult environment. It is partially observed, noisy, and non-stationary, there may be outliers and regimes.

Time also plays a key role and decisions again may have long term consequences. In contradistinction to standard methods we are not looking to apply single period prediction and then combine these predictions using an optimiser.

This is akin to supervised learning, but in the real world our actions may have long term effects and indeed actions taken by our agents may be reacted against by the environment. Most standard methods are single period and represent a two stage process, this involves two sets of parameters and forecast error is not utility, so we may even be optimising the wrong target.

Other works give up upon some of our ability to predict and are thus more heuristic but more practical methods for allocation decisions, albeit pessimistic. An information bottleneck is created between the supervised forecast error minimisation and the subsequent forecasts which are then used by an optimiser some argue that this also serves as an error maximiser and indeed has its own parameters to be found.

## Comments