Neural Networks in Trading

Yes, you will be awarded with a certification from QuantInsti after successfully completing the online learning units.

No, there are no live or classroom sessions in the course. You can ask your queries on community and get responses from fellow learners and faculty members.

Yes, you can ask your queries related to the course on the community: https://quantra.quantinsti.com/community

Fast-speed internet connection and a browser application are required for this course. For best experience, use Chrome.

There is no admission criterion. You are recommended to go through the prerequisites section and be aware of skill sets gained and required to learn most from the course.

We respect your time, and hence, we offer concise but effective short-term courses created under professional guidance. We try to offer the most value within the shortest time. There are a few courses on Quantra which are free of cost. Please check the price of the course before enrolling in it. Once a purchase is made, we offer complete course content. For paid courses, we follow a 'no refund' policy.

Some of the course material is downloadable such as Python notebooks with strategy codes. We also guide you how to use these codes on your own system to practice further.

We focus on teaching these quantitative and machine learning techniques and how learners can use them for developing their own strategies. You may or may not be able to directly use them in your own system. Please do note that we are not advising or offering any trading/investment services. The strategies are used for learning & understanding purposes and we don't take any responsibility for the performance or any profit or losses that using these techniques results in.

Quantra environment is a zero-installation solution to get beginners to start off with coding in Python. While learning you won't have to download or install anything! However, if you wish to later implement the learning on your system, you can definitely do that. All the notebooks in the Quantra portal are available for download at the end of each course and they can be run in the local system just the same as they run in the portal. The user can modify/tweak/rework all such code files as per his need. We encourage you to implement different concepts learnt from different learning tracks into your trading strategy to make it more suited to the real-world scenario.

No. We provide you guidance on how to create strategy using different techniques and indicators, but no strategy is plug and play. A lot of effort is required to backtest any strategy, after which we fine-tune the strategy parameters and see the performance on paper trading before we finally implement the live execution of trades.

Lifetime access means that once you enroll in the course, you will have unlimited access to all course materials, including videos, resources, readings, and other learning materials for as long as the course remains available online. There are no time limits or expiration dates on your access, allowing you to learn at your own pace and revisit the content whenever you need it, even after you've completed the course. It's important to note that "lifetime" refers to the lifetime of the course itself—if the platform or course is discontinued for any reason, we will inform you in advance. This will allow you enough time to download or access any course materials you need for future use.

Neural Networks are a plethora of architectures. Simple feed-forward networks and Recurrent Neural Networks (RNN) are generally the ones used with Time Series data.

Feed-Forward Networks takes in a window of historical prices all at once and predict a single price or the direction of market movement for one day in the future

Similarly, RNN takes in a window of prices sequentially and makes a similar sort of prediction as a feed-forward network.

Neural Networks are state-of-the-art predictors. They do very well in identifying non-linear patterns in time-series data. As a consequence, they can outperform manual technical analysis and traditional statistical methods in identifying trends, momentums, seasonalities etc. even in short terms. Hence, they are increasingly being used by institutional and retail investors.

A fully connected Neural Network is one where each neuron in a given layer is connected to every other neuron in the subsequent layer. These neurons send weighted 'signals' to neurons in the subsequent layers. These 'signals' in the final/output layers lead to predictions. 'signals' are sent back based on the accuracy of these predictions. These reverse 'signals' are used to change the weights of these neurons. This leads to learning.
Fully connected Neural Networks are covered in various sections of this course.

Multilayer Perceptrons or MLPs are the classical feed-forward Neural Networks. They have multiple layers. One input layer, multiple hidden layers and one output. Each layer is made up of multiple neurons which pass signals to the neurons of the subsequent layers. They also give feedback to neurons in the layer before them. This feedback helps in fine-tuning to the patterns in the data. This fine-tuning is the learning that happens in these models. MLPs are suitable for the classification problems where the aim is to predict labels or classes based on a given input. They are also suitable for regression problems where a real-valued quantity is predicted given a set of inputs.
MLPs are covered in section 3 of this course.

Backpropagation is an algorithm used by Neural Networks. This algorithm calculates the gradients of weight parameters of Neural Networks with respect to the error or loss function. The calculation happens going back. It is calculated for weights across all layers of the network. This gradient for each weight with respect to the loss function is used to change the weights such that the total error or loss over the dataset minimises. This is how 'learning' happens. Backpropagation makes use of the derivative chain rule to calculate gradients.
Backpropagation is explained in section 4 of this course.

Yes, Neural Networks can be used for classification. Classification is the categorisation of a given input by a model to a set of categories or classes. They can also be used as predictors in a variety of other problems like regression and agent-based modelling.
Classification using Neural Networks is covered in section 3 and 4 of this course.

Keras is a high-level neural network API. It is written in Python and is capable of using many back-end neural network computation engines like Tensorflow etc. It provides many high-level layers like Dense, RNNs, CNNs etc. as features.
It is covered in various sections of this course.