jjcap

Friday, September 27, 2019

Return to main page

• Dean explaining some of the fundamental formulas/ratios he got from different tech companies.
  1. Quick Ratio
  2. Debt-to-Equity Ratio = (total debt/ total equity)
  3. Profitability Ratios
  4. Price/Sales ratio
• Should make a requirements.py file

Adeet’s Presentation

Classifying Directional Stock Price Movement

• Machine Learning - Computers being able to recognize patterns in a data set and then making predictions through the training/fit for that data set. You would then take that model and apply it to different data sets.

• Overfit - model is over compensating or learning pattern too well, and applying a specific pattern and in general that will not be the case.

• It will predict either 1 or 0
  • 1 meaning price will increase the next day (upwards movement)
  • 0 meaning it will stay the same or decrease (stagnant/downwards movement)

• Random Walk theory - maximum prediction accuracy at 50%

• Efficient market hypothesis only good for long trends, not the best at predicting short spikes or trends.

• Main Problem:
  • Two components
    • The information and data sources to gather
    • The information/data you want to feed into the algorithm

• Feature = another name for a variable

• Feature Testing: testing different variables to achieve the target.

• Doesn’t use price data because of multicollinearity.

• Two variables would be very close/similar together and it could offset our best predictions from the model.

• For the indicators: accuracy in machine learning isn’t the best indicator to go buy.

• Precision and Recall are closely related

• Ensemble models could be the way to go. Basically taking a bunch of weak models and finding the averages between them in the feature testing, and trying to combine into a stronger model.