![sri palani's cover illustration](/_next/image?url=%2F_next%2Fstatic%2Fmedia%2Fdefault-cover.df740e10.webp&w=3840&q=75)
![sri palani's profile](/_next/image?url=%2F_next%2Fstatic%2Fmedia%2Fgeneric-profile.ea743bf4.webp&w=640&q=75)
sri palani
Class of 2025
About
Projects
- "Can we accurately predict short term price action of a company by combining fundamental analysis with macroeconomic trends?" with mentor Vijay (July 14, 2024)
Project Portfolio
Can we accurately predict short term price action of a company by combining fundamental analysis with macroeconomic trends?
Started Jan. 25, 2024
Abstract or project description
Methodology:
Find 100 companies on the NASDAQ/S&P500 (or similar industrial complex index fund). Company specific data: Aggregate key fundamentals: balance sheet, 10k, etc Subgroups based on: Sector, Industry, Market Cap (mega-cap, large-cap, mid-cap, small-cap) Fundamentals: Share price, market cap, PE/PEG, P/B, P/S, Net Income/Revenue, FCF, maybe Dividends, Assets/Debt/Liabilities, EBITDA Notes: Connect sheet with Google Finance API so that the sheet auto updates.
Develop scoring metric for the general state of the macroeconomy (some weighted average i.e -1 to 1). Variables involved: geopolitical climate, consumer spending, presidential party in office, general state of healthcare & climate. Notes: Use the consumer sentiment score and then create a AI that will use a DCF formula and predict the stock price action in the next couple weeks
Start 6-12 months back (or YTD) to backtest and train model. Ex. Train on 9 months, test on 3 months. Step 1: Create a massive dataframe on pandas which effectively contains one excel sheet worth of information for each month March 1st 2023 to March 1st 2024 (12 sheets). Add the change in each metric between month to month.
Step 2: In a separate column track the rate of change of consumer sentiment score for each of those months as well (March 1st 2023-March 1st 2024)
Step 3: The goal is to predict the price change.
Step 4: Create a initial classifier model: -Train/Test split: 70/30 -Use 70% of the training data to predict 30% of the remaining training data Step 5: Fine tune by removing unnecessary/weakly correlated variables, adjust weightages of model parameters, etc. until we start seeing a prediction accuracy of 90%