Sentiment based Stock Analysis

Objective The project aims to develop a sentiment analysis model to predict the movement of stock prices based on textual data from news articles. Coding Languages Used: Python (Librararies Used: Numpy, Pandas, vaderSentiment, TextBlob, sklearn, imblearn, backtesting, yfinance ) Procedure

1. News article headlines scraping using New York Times API key. (from 1 Jan 2023 to 10 April 2024)I have extracted data for Apple, Nvidia and Tesla.
2. Stock Price Extraction(OHLC) using yfinance.
3. Labelling: (i). Increase(if price change>0.01) (ii). Decrease (if price change<-0.01) else no change.
4. Horizontally concatenate stock price and news headline data for the dates present in both of them.
5. Replacing Categorical Labels(‘Increase’:1 , ‘Decrease’: -1 and ‘no_change’:0 ) 6.Applied stock_sentiment_analyser to headline data and got compound, positive, negative and neutral scores.
6. Getting subjectivity and polarity scores for headlines using TextBlob library.
7. Applying Train_Test_split to distribute all three dataframes(one for each company) into 80:20 to get training: testing data.
8. Applying CountVectorizer().fit_transform to headlines in training and CountVectorizer().transform to headlines in testing data.
9. Concatenating training parts for each dataframe and then concatenating testing parts.
10. Balance imbalance in Training data target column i.e. Label using SMOTE.
11. Using Random Forest, SVM and Logistic Regression to train the model on training data.
12. Evaluate model using accuracy score and Classification report.
13. developed an strategy use backtesting to backtest it only on Tesla Data.
14. Using bt.plot() to plot three graphs:
    (i). First graph tells about the net equity I have. (ii). Second tells about the profit and loss. It also contains candlestick-patterns. (iii). Third graph tells about the volume traded oon each date. Strategy
15. Calculating SMA(Simple Moving Average) :SMAs are used to smooth out price data and identify trends by averaging out short-term fluctuations. Two SMAs with different periods (4 and 8) are used to detect trend changes through crossover signals.
16. Sentiment threshold is used to filter sentiment scores, ensuring that only strong sentiment signals (either positive or negative) trigger trades.
17. Uses sentiment analysis scores (‘Compound’, ‘Polarity’, ‘Subjectivity’) to decide if the market sentiment supports a 'long' or 'short' position. (i). subjectivity indicates the presence of personal views, opinions, or feelings. (ii). The polarity score measures the emotional tone, indicating whether the expressed sentiment is positive, negative, or neutral. (iii). Compound score combines the effects of individual sentiment-bearing words into a single score.It captures the overall sentiment in a straightforward manner.
18. Bullish Crossover: When the shorter-period SMA (sma1) crosses above the longer-period SMA (sma2), it indicates a potential uptrend, generating a 'buy' signal. Bearish Crossover: When the shorter-period SMA (sma1) crosses below the longer-period SMA (sma2), it indicates a potential downtrend, generating a 'sell' signal.
19. Positive Sentiment (Long): If the subjectivity is high (greater than 0.5) and the compound score is positive (greater than 0.1), it generates a 'long' signal. Negative Sentiment (Short): If the subjectivity is low (less than 0.5) and the compound score is negative (less than -0.1), it generates a 'short' signal.
20. Buy Signal: If an SMA bullish crossover occurs and the sentiment signal is 'long', the strategy closes any existing short positions and opens a long position. Sell Signal: If an SMA bearish crossover occurs and the sentiment signal is 'short', the strategy closes any existing long positions and opens a short position. No Trade: If there is no sentiment signal, the strategy
    closes any open positions and remains out of the market.

Strategy Analysis:

1. The strategy has a Sharpe Ratio of approximately 0.634, indicating a good risk-adjusted return compared to the risk-free rate.
2. Max Drawdown: The maximum drawdown experienced during the backtest period is around -7.178%, which means that strategy does not expose the portfolio to excessive risk.
3. The strategy made a total of 9 trades over the backtesting period. This low number of trades suggests the strategy is relatively selective
4. A win rate of 55.55% indicates that slightly more than half of the trades were successful, which is a positive outcome.
5. A value above 1 indicates that the strategy is profitable, and a value of 1.84234 is quite good, suggesting that the strategy generates almost twice as much profit as it incurs in losses. Conclusion: A high risk appetite trader will be disappointed by the strategy made as it has primary focus on preventing losses but the strategy will be highly beneficiary for a person who wants to lower the risk as well as to the beginner who wants to build the confidence in share market. Overall, the strategy does relatively lesser number of trades and also generates only mediocre returns which indicates its risk-free attitude. As most of the people in India are middle and lower-middle class, the strategy is highly suited for them.