Forecast Stock Price Using Amazon SageMaker Canvas

Agenda

1. Overview
2. Import the dataset in Canvas
3. Building and Training a ML model
4. Using the model to generate predictions

Overview

In this lab, you will assume the role of a financial analyst working for a stock investing institution. You will use historical stock price for a fictitious company XXYZ, dated between 2016 and 2020, to build a forecast model, and use it to predict future stock price. For those interested, a jupyter notebook is provided for the construction of the simulated dataset here.

DISCLAIMER. The exercise is to merely walk through the steps required to build a forecast model against simulated time series data. Investors should consider investing only if they can accept the risks associated with investing including a loss of invested capital.

For purpose of this lab, the data schema is defined as follows. The stock symbol is stored as "Item_Id" , trading date as "Date", and closing price as "Close" for easy reference. You may download the CSV file before proceeding.

In terms of AWS environment, please ensure you have SageMaker Canvas properly configured. You may reference official documentation to set up the resource. Once ready, you can go to the AWS Management Console, search S3 in the searchbox, and go to S3 service console.

In the S3 console, click on the sagemaker-studio-* bucket.

The sagemaker-studio-* bucket was created automatically when you created the SageMaker Studio domain.

Click Upload.

On the Upload page, drag and drop the daily_close_price.csv file you've just downloaded, then click Upload at the bottom of the page. Once the upload is complete, you can click the top-right Close button. You should see the file uploaded to your bucket.

Import the dataset in Canvas

Go to the SageMaker Canvas. On the left menu, you can click the second icon to navigate to the Datasets section, then click the Import button.

Now, select the bucket where we've previously uploaded our dataset, the sagemaker-studio-* bucket.

You can now select the daily_close_price.csv file by selecting the checkbox. Two new buttons will pop-up at the bottom of your page: Preview all and Import Data. We can choose to preview.

You now preview a 100-rows of the dataset you're about to import. Once you're done checking that it's indeed the right dataset, you can click on Import Data.

Building and Training a ML model

Now that the dataset is imported, you can create a new model by going to the Models screen, and clicking on the + New Model button.

On the Create new model popup screen, write store_sales_forecast_model for the model name and click on the Create button.

On the Select dataset screen, select daily_price_close_only for the dataset and click on the Select dataset button.

On the next screen, you can configure the model type for training. You can also select columns to see statistics of each column. Select Close for the Target column field. Canvas will automatically select Time series forecasting as the model type. Click on the Configure link.

The Time Series Forecasting configuration popup screen, you are asked to provide a few information:

• The items field: how you identify your items in the datasets. For this use case, select Item_Id since we are planning to forecast price per stock symbol
• The group column: if you have logical groupings of the items selected above. You can choose the feature here. We don't have one for this use case, but examples group might be NYSE, Nasdaq, or other groupings of stock symbols as your use case desires.
• The time stamp field: select Date, which is the feature that contains the time stamp information. Canvas supports data timestamp in the format YYYY-MM-DD (e.g.: 2018-01-01). You may reference other supported data formats here.
• Write 30 in the number of Days field, assuming we want to forecast stock closing price 30 days into the future.
• Select United States as the country for holidays field, as additional information to feed into model training.

Finally, click on the Save button.

Now that the configuration is done, we're ready to train the model. Currently SageMaker Canvas does not support Quick Build for Time-Series Forecasting. We will select the Standard Build option. Model will take around 2-4 hours to train.

Using the model to generate predictions

When the model training finishes, you will be routed to the Analyze tab. There, you can see the average prediction accuracy, and the column impact on prediction outcome. Click on the Predict button, and prepare to run some forecast.

In order to create forecast predictions, you have to provide first the date range for which the forecast prediction can be made. Then, you can generate forecast predictions for all the items in the dataset or a specific item.

In our workshop, we choose the Single item option, and select XXYZ from the dropdown list, and Canvas generates a prediction for our item, showing the average prediction, an upper bound and a lower bound.

For the generated forecast prediction, you can click on the Download dropdown menu button to download the forecast prediction chart as image or forecast prediction values as CSV file.

Congratulations! You've completed the lab to forecast stock price using SageMaker Canvas.