/fin-ai

Primary LanguageJupyter Notebook

sharpe.ipynb

This code is performing a Monte Carlo simulation to find the optimal weights of a portfolio consisting of four stocks (AAPL, NVDA, AMD, and IBM) based on their historical closing prices from 2018 onwards. The objective is to maximize the Sharpe ratio, which is a measure of risk-adjusted return.

Here's a step-by-step interpretation and commentary on the code:

  1. Data Retrieval:

    • The code uses the yfinance library to fetch historical stock data for AAPL, NVDA, AMD, and IBM starting from '2018-01-01'.

  2. Data Organization:

    • The fetched data for each stock is concatenated side-by-side using pd.concat().
    • The columns are renamed using a multi-index to distinguish between the stocks and their respective data columns (Open, High, Low, etc.).
    • Log returns for the 'Close' prices of each stock are calculated using the formula log(price_today/price_yesterday), and the first NaN value is dropped.

  3. Monte Carlo Simulation:

    • The code simulates 6,000 portfolios with random weights for the four stocks.
    • For each portfolio, it calculates the expected annual return, expected annual volatility, and the Sharpe ratio.
    • The weights of the assets in each portfolio are stored in all_weights.

  4. Results Display:

    • The code prints the maximum and minimum Sharpe ratios from the simulation and the index (location) of the portfolio with the maximum Sharpe ratio.
    • It also prints the weights of the assets in the portfolio with the maximum Sharpe ratio.

  5. Visualization:

    • A scatter plot is created where each point represents a simulated portfolio. The x-axis represents the portfolio's volatility, and the y-axis represents its return. The color of each point is determined by its Sharpe ratio, with warmer colors indicating higher Sharpe ratios.
    • The portfolio with the maximum Sharpe ratio is highlighted with a red dot.

Comments on the Output:

  • The code aims to find the optimal mix of the four stocks to maximize the Sharpe ratio. The portfolio with the highest Sharpe ratio is considered the most desirable because it provides the highest return for a given level of risk (or the lowest risk for a given level of return).

  • The scatter plot provides a visual representation of the risk-return trade-off for the various portfolio combinations. The "efficient frontier" can be visualized as the upper boundary of the scatter plot, where you get the highest return for a given level of risk.

  • The red dot on the scatter plot indicates the portfolio with the highest Sharpe ratio. This is the point where you get the maximum return per unit of risk.

  • The printed weights corresponding to the maximum Sharpe ratio provide an actionable insight: if an investor wants to achieve the highest risk-adjusted return based on historical data, they should allocate their capital according to these weights among AAPL, NVDA, AMD, and IBM.

  • It's important to note that this analysis is based on historical data, and past performance is not indicative of future results. The optimal portfolio based on past data might not be optimal in the future. Regular rebalancing and consideration of other factors (like transaction costs, taxes, and fundamental analysis) are essential for practical portfolio management.

main.ipynb

This code is attempting to predict the price of Ethereum (ETH) using a Long Short-Term Memory (LSTM) neural network. LSTMs are a type of recurrent neural network (RNN) that are particularly well-suited for sequence prediction problems, such as time series forecasting. Here's a step-by-step interpretation and commentary on the code:

  1. Data Retrieval:

    • The code fetches historical data for Ethereum (ETH) from Yahoo Finance starting from '2018-01-01' using the yfinance library.

  2. Data Preparation:

    • The code extracts the 'Open' and 'High' prices from the Ethereum data for training.
    • The data is normalized using the MinMaxScaler from Scikit-Learn, which scales the data to be between 0 and 1. This is a common preprocessing step for neural networks to help improve convergence during training.

  3. Feature Engineering:

    • The code constructs the training dataset such that for each day, it uses the previous 60 days' 'Open' prices to predict the next day's 'Open' price.

  4. Model Building:

    • An LSTM-based neural network model is constructed using the Keras library.
    • The model consists of four LSTM layers, each followed by a dropout layer to prevent overfitting. The final layer is a dense layer with a single neuron to predict the 'Open' price.
    • The model is compiled using the Adam optimizer and mean squared error as the loss function.

  5. Model Training:

    • The LSTM model is trained on the prepared training data for 100 epochs with a batch size of 32.

  6. Price Prediction:

    • The code prepares the test dataset in a similar manner as the training dataset.
    • The trained LSTM model is then used to predict the Ethereum prices.
    • The predicted prices are inverse transformed to get them back to their original scale.

  7. Visualization:

    • A plot is generated that compares the real Ethereum prices with the predicted prices.

Comments on the Output:

  • The LSTM model aims to capture the temporal dependencies in the Ethereum price data to make future predictions.

  • The visualization will show how well the LSTM model's predictions align with the actual Ethereum prices. If the predicted line closely follows the real price line, it indicates that the model has learned meaningful patterns from the historical data.

  • The code have a few issues:

    • The variable training_data_df is used but not defined. It seems like it should be training_data.
    • The variable dataset_test is referenced but not defined.
    • The variable real_stock_price should probably be real_price.
    • The variable predicted_price should probably be predicted_stock_price.
    • The model is used for prediction before it's defined and trained. The prediction code block should be moved after the model training block.

  • It's important to note that predicting stock or cryptocurrency prices is a challenging task. Even if the model performs well on historical data, it doesn't guarantee that it will perform well on future unseen data. External factors, news, and market sentiment can influence prices, and these factors might not be captured by historical price data alone.

  • Regular evaluation and possibly retraining of the model are essential for practical applications.