Tonic's picture
improve regime detection
542229d unverified
import gradio as gr
import pandas as pd
import numpy as np
from datetime import datetime, timedelta
import yfinance as yf
import torch
from chronos import ChronosPipeline
import plotly.graph_objects as go
from plotly.subplots import make_subplots
from sklearn.preprocessing import MinMaxScaler
import plotly.express as px
from typing import Dict, List, Tuple, Optional
import json
import spaces
import gc
import pytz
import time
import random
from scipy import stats
from scipy.optimize import minimize
import warnings
warnings.filterwarnings('ignore')
# Additional imports for advanced features
try:
from hmmlearn import hmm
HMM_AVAILABLE = True
except ImportError:
HMM_AVAILABLE = False
print("Warning: hmmlearn not available. Regime detection will use simplified methods.")
try:
from sklearn.ensemble import RandomForestRegressor
from sklearn.linear_model import LinearRegression
ENSEMBLE_AVAILABLE = True
except ImportError:
ENSEMBLE_AVAILABLE = False
print("Warning: scikit-learn not available. Ensemble methods will be simplified.")
# Initialize global variables
pipeline = None
scaler = MinMaxScaler(feature_range=(-1, 1))
scaler.fit_transform([[-1, 1]])
# Global market data cache
market_data_cache = {}
cache_expiry = {}
CACHE_DURATION = 3600 # 1 hour cache
def retry_yfinance_request(func, max_retries=3, initial_delay=1):
"""
Retry mechanism for yfinance requests with exponential backoff.
Args:
func: Function to retry
max_retries: Maximum number of retry attempts
initial_delay: Initial delay in seconds before first retry
Returns:
Result of the function call if successful
"""
for attempt in range(max_retries):
try:
return func()
except Exception as e:
if "401" in str(e) and attempt < max_retries - 1:
# Calculate delay with exponential backoff and jitter
delay = initial_delay * (2 ** attempt) + random.uniform(0, 1)
time.sleep(delay)
continue
raise e
def clear_gpu_memory():
"""Clear GPU memory cache"""
if torch.cuda.is_available():
torch.cuda.empty_cache()
gc.collect()
@spaces.GPU()
def load_pipeline():
"""Load the Chronos model without GPU configuration"""
global pipeline
try:
if pipeline is None:
clear_gpu_memory()
print("Loading Chronos model...")
pipeline = ChronosPipeline.from_pretrained(
"amazon/chronos-t5-large",
device_map="cuda", # Force CUDA device mapping
torch_dtype=torch.float16,
low_cpu_mem_usage=True,
trust_remote_code=True,
use_safetensors=True
)
# Set model to evaluation mode
pipeline.model = pipeline.model.eval()
# Disable gradient computation
for param in pipeline.model.parameters():
param.requires_grad = False
print("Chronos model loaded successfully")
return pipeline
except Exception as e:
print(f"Error loading pipeline: {str(e)}")
print(f"Error type: {type(e)}")
print(f"Error details: {str(e)}")
raise RuntimeError(f"Failed to load model: {str(e)}")
def is_market_open() -> bool:
"""Check if the market is currently open"""
now = datetime.now()
# Check if it's a weekday (0 = Monday, 6 = Sunday)
if now.weekday() >= 5: # Saturday or Sunday
return False
# Check if it's during market hours (9:30 AM - 4:00 PM ET)
et_time = now.astimezone(pytz.timezone('US/Eastern'))
market_open = et_time.replace(hour=9, minute=30, second=0, microsecond=0)
market_close = et_time.replace(hour=16, minute=0, second=0, microsecond=0)
return market_open <= et_time <= market_close
def get_next_trading_day() -> datetime:
"""Get the next trading day"""
now = datetime.now()
next_day = now + timedelta(days=1)
# Skip weekends
while next_day.weekday() >= 5: # Saturday or Sunday
next_day += timedelta(days=1)
return next_day
def get_historical_data(symbol: str, timeframe: str = "1d", lookback_days: int = 365) -> pd.DataFrame:
"""
Fetch historical data using yfinance with enhanced support for intraday data.
Args:
symbol (str): The stock symbol (e.g., 'AAPL')
timeframe (str): The timeframe for data ('1d', '1h', '15m')
lookback_days (int): Number of days to look back
Returns:
pd.DataFrame: Historical data with OHLCV and technical indicators
"""
try:
# Check if market is open for intraday data
if timeframe in ["1h", "15m"] and not is_market_open():
next_trading_day = get_next_trading_day()
raise Exception(f"Market is currently closed. Next trading day is {next_trading_day.strftime('%Y-%m-%d')}")
# Map timeframe to yfinance interval and adjust lookback period
tf_map = {
"1d": "1d",
"1h": "1h",
"15m": "15m"
}
interval = tf_map.get(timeframe, "1d")
# Adjust lookback period based on timeframe and yfinance limits
if timeframe == "1h":
lookback_days = min(lookback_days, 60) # Yahoo allows up to 60 days for hourly data
elif timeframe == "15m":
lookback_days = min(lookback_days, 7) # Yahoo allows up to 7 days for 15m data
# Calculate date range
end_date = datetime.now()
start_date = end_date - timedelta(days=lookback_days)
# Fetch data using yfinance with retry mechanism
ticker = yf.Ticker(symbol)
def fetch_history():
return ticker.history(
start=start_date,
end=end_date,
interval=interval,
prepost=True, # Include pre/post market data for intraday
actions=True, # Include dividends and splits
auto_adjust=True, # Automatically adjust for splits
back_adjust=True, # Back-adjust data for splits
repair=True # Repair missing data points
)
df = retry_yfinance_request(fetch_history)
if df.empty:
raise Exception(f"No data available for {symbol} in {timeframe} timeframe")
# Ensure all required columns are present and numeric
required_columns = ['Open', 'High', 'Low', 'Close', 'Volume']
for col in required_columns:
if col not in df.columns:
raise Exception(f"Missing required column: {col}")
df[col] = pd.to_numeric(df[col], errors='coerce')
# Get additional info for structured products with retry mechanism
def fetch_info():
info = ticker.info
if info is None:
raise Exception(f"Could not fetch company info for {symbol}")
return info
try:
info = retry_yfinance_request(fetch_info)
df['Market_Cap'] = float(info.get('marketCap', 0))
df['Sector'] = info.get('sector', 'Unknown')
df['Industry'] = info.get('industry', 'Unknown')
df['Dividend_Yield'] = float(info.get('dividendYield', 0))
# Add additional company metrics
df['Enterprise_Value'] = float(info.get('enterpriseValue', 0))
df['P/E_Ratio'] = float(info.get('trailingPE', 0))
df['Forward_P/E'] = float(info.get('forwardPE', 0))
df['PEG_Ratio'] = float(info.get('pegRatio', 0))
df['Price_to_Book'] = float(info.get('priceToBook', 0))
df['Price_to_Sales'] = float(info.get('priceToSalesTrailing12Months', 0))
df['Return_on_Equity'] = float(info.get('returnOnEquity', 0))
df['Return_on_Assets'] = float(info.get('returnOnAssets', 0))
df['Debt_to_Equity'] = float(info.get('debtToEquity', 0))
df['Current_Ratio'] = float(info.get('currentRatio', 0))
df['Quick_Ratio'] = float(info.get('quickRatio', 0))
df['Gross_Margin'] = float(info.get('grossMargins', 0))
df['Operating_Margin'] = float(info.get('operatingMargins', 0))
df['Net_Margin'] = float(info.get('netIncomeToCommon', 0))
except Exception as e:
print(f"Warning: Could not fetch company info for {symbol}: {str(e)}")
# Set default values for missing info
df['Market_Cap'] = 0.0
df['Sector'] = 'Unknown'
df['Industry'] = 'Unknown'
df['Dividend_Yield'] = 0.0
df['Enterprise_Value'] = 0.0
df['P/E_Ratio'] = 0.0
df['Forward_P/E'] = 0.0
df['PEG_Ratio'] = 0.0
df['Price_to_Book'] = 0.0
df['Price_to_Sales'] = 0.0
df['Return_on_Equity'] = 0.0
df['Return_on_Assets'] = 0.0
df['Debt_to_Equity'] = 0.0
df['Current_Ratio'] = 0.0
df['Quick_Ratio'] = 0.0
df['Gross_Margin'] = 0.0
df['Operating_Margin'] = 0.0
df['Net_Margin'] = 0.0
# Calculate technical indicators with adjusted windows based on timeframe
if timeframe == "1d":
sma_window_20 = 20
sma_window_50 = 50
sma_window_200 = 200
vol_window = 20
elif timeframe == "1h":
sma_window_20 = 20 * 6 # 5 trading days
sma_window_50 = 50 * 6 # ~10 trading days
sma_window_200 = 200 * 6 # ~40 trading days
vol_window = 20 * 6
else: # 15m
sma_window_20 = 20 * 24 # 5 trading days
sma_window_50 = 50 * 24 # ~10 trading days
sma_window_200 = 200 * 24 # ~40 trading days
vol_window = 20 * 24
# Calculate technical indicators
df['SMA_20'] = df['Close'].rolling(window=sma_window_20, min_periods=1).mean()
df['SMA_50'] = df['Close'].rolling(window=sma_window_50, min_periods=1).mean()
df['SMA_200'] = df['Close'].rolling(window=sma_window_200, min_periods=1).mean()
df['RSI'] = calculate_rsi(df['Close'])
df['MACD'], df['MACD_Signal'] = calculate_macd(df['Close'])
df['BB_Upper'], df['BB_Middle'], df['BB_Lower'] = calculate_bollinger_bands(df['Close'])
# Calculate returns and volatility
df['Returns'] = df['Close'].pct_change()
df['Volatility'] = df['Returns'].rolling(window=vol_window, min_periods=1).std()
df['Annualized_Vol'] = df['Volatility'] * np.sqrt(252)
# Calculate drawdown metrics
df['Rolling_Max'] = df['Close'].rolling(window=len(df), min_periods=1).max()
df['Drawdown'] = (df['Close'] - df['Rolling_Max']) / df['Rolling_Max']
df['Max_Drawdown'] = df['Drawdown'].rolling(window=len(df), min_periods=1).min()
# Calculate liquidity metrics
df['Avg_Daily_Volume'] = df['Volume'].rolling(window=vol_window, min_periods=1).mean()
df['Volume_Volatility'] = df['Volume'].rolling(window=vol_window, min_periods=1).std()
# Calculate additional intraday metrics for shorter timeframes
if timeframe in ["1h", "15m"]:
# Intraday volatility
df['Intraday_High_Low'] = (df['High'] - df['Low']) / df['Close']
df['Intraday_Volatility'] = df['Intraday_High_Low'].rolling(window=vol_window, min_periods=1).mean()
# Volume analysis
df['Volume_Price_Trend'] = (df['Volume'] * df['Returns']).rolling(window=vol_window, min_periods=1).sum()
df['Volume_SMA'] = df['Volume'].rolling(window=vol_window, min_periods=1).mean()
df['Volume_Ratio'] = df['Volume'] / df['Volume_SMA']
# Price momentum
df['Price_Momentum'] = df['Close'].pct_change(periods=5)
df['Volume_Momentum'] = df['Volume'].pct_change(periods=5)
# Fill NaN values using forward fill then backward fill
df = df.ffill().bfill()
# Ensure we have enough data points
min_required_points = 64 # Minimum required for Chronos
if len(df) < min_required_points:
# Try to fetch more historical data with retry mechanism
extended_start_date = start_date - timedelta(days=min_required_points - len(df))
def fetch_extended_history():
return ticker.history(
start=extended_start_date,
end=start_date,
interval=interval,
prepost=True,
actions=True,
auto_adjust=True,
back_adjust=True,
repair=True
)
extended_df = retry_yfinance_request(fetch_extended_history)
if not extended_df.empty:
df = pd.concat([extended_df, df])
df = df.ffill().bfill()
if len(df) < 2:
raise Exception(f"Insufficient data points for {symbol} in {timeframe} timeframe")
# Final check for any remaining None values
df = df.fillna(0)
return df
except Exception as e:
raise Exception(f"Error fetching historical data for {symbol}: {str(e)}")
def calculate_rsi(prices: pd.Series, period: int = 14) -> pd.Series:
"""Calculate Relative Strength Index"""
# Handle None values by forward filling
prices = prices.ffill().bfill()
delta = prices.diff()
gain = (delta.where(delta > 0, 0)).rolling(window=period).mean()
loss = (-delta.where(delta < 0, 0)).rolling(window=period).mean()
rs = gain / loss
return 100 - (100 / (1 + rs))
def calculate_macd(prices: pd.Series, fast: int = 12, slow: int = 26, signal: int = 9) -> Tuple[pd.Series, pd.Series]:
"""Calculate MACD and Signal line"""
# Handle None values by forward filling
prices = prices.ffill().bfill()
exp1 = prices.ewm(span=fast, adjust=False).mean()
exp2 = prices.ewm(span=slow, adjust=False).mean()
macd = exp1 - exp2
signal_line = macd.ewm(span=signal, adjust=False).mean()
return macd, signal_line
def calculate_bollinger_bands(prices: pd.Series, period: int = 20, std_dev: int = 2) -> Tuple[pd.Series, pd.Series, pd.Series]:
"""Calculate Bollinger Bands"""
# Handle None values by forward filling
prices = prices.ffill().bfill()
middle_band = prices.rolling(window=period).mean()
std = prices.rolling(window=period).std()
upper_band = middle_band + (std * std_dev)
lower_band = middle_band - (std * std_dev)
return upper_band, middle_band, lower_band
@spaces.GPU()
def make_prediction(symbol: str, timeframe: str = "1d", prediction_days: int = 5, strategy: str = "chronos",
use_ensemble: bool = True, use_regime_detection: bool = True, use_stress_testing: bool = True,
risk_free_rate: float = 0.02, ensemble_weights: Dict = None,
market_index: str = "^GSPC",
random_real_points: int = 4, use_smoothing: bool = True,
smoothing_type: str = "exponential", smoothing_window: int = 5,
smoothing_alpha: float = 0.3) -> Tuple[Dict, go.Figure]:
"""
Make prediction using selected strategy with advanced features.
Args:
symbol (str): Stock symbol
timeframe (str): Data timeframe ('1d', '1h', '15m')
prediction_days (int): Number of days to predict
strategy (str): Prediction strategy to use
use_ensemble (bool): Whether to use ensemble methods
use_regime_detection (bool): Whether to use regime detection
use_stress_testing (bool): Whether to perform stress testing
risk_free_rate (float): Risk-free rate for calculations
ensemble_weights (Dict): Weights for ensemble models
market_index (str): Market index for correlation analysis
random_real_points (int): Number of random real points to include in long-horizon context
use_smoothing (bool): Whether to apply smoothing to predictions
smoothing_type (str): Type of smoothing to apply ('exponential', 'moving_average', 'kalman', 'savitzky_golay', 'none')
Returns:
Tuple[Dict, go.Figure]: Trading signals and visualization plot
"""
try:
# Get historical data
df = get_historical_data(symbol, timeframe)
if strategy == "chronos":
try:
# Prepare data for Chronos
prices = df['Close'].values
chronos_context_size = 64 # Chronos model's context window size (fixed at 64)
input_context_size = len(prices) # Available input data can be much larger
# Use a larger range for scaler fitting to get better normalization
scaler_range = min(input_context_size, chronos_context_size * 2) # Use up to 128 points for scaler
# Select the most recent chronos_context_size points for the model input
context_window = prices[-chronos_context_size:]
scaler = MinMaxScaler(feature_range=(-1, 1))
# Fit scaler on a larger range for better normalization
scaler.fit(prices[-scaler_range:].reshape(-1, 1))
normalized_prices = scaler.transform(context_window.reshape(-1, 1)).flatten()
# Ensure we have enough data points for Chronos
min_data_points = chronos_context_size
if len(normalized_prices) < min_data_points:
padding = np.full(min_data_points - len(normalized_prices), normalized_prices[-1])
normalized_prices = np.concatenate([padding, normalized_prices])
elif len(normalized_prices) > min_data_points:
normalized_prices = normalized_prices[-min_data_points:]
# Load pipeline and move to GPU
pipe = load_pipeline()
# Get the model's device and dtype
device = torch.device("cuda:0") # Force CUDA device
dtype = torch.float16 # Force float16
print(f"Model device: {device}")
print(f"Model dtype: {dtype}")
# Convert to tensor and ensure proper shape and device
context = torch.tensor(normalized_prices, dtype=dtype, device=device)
# Adjust prediction length based on timeframe
if timeframe == "1d":
max_prediction_length = chronos_context_size # 64 days
actual_prediction_length = min(prediction_days, max_prediction_length)
trim_length = prediction_days
elif timeframe == "1h":
max_prediction_length = chronos_context_size # 64 hours
actual_prediction_length = min(prediction_days * 24, max_prediction_length)
trim_length = prediction_days * 24
else: # 15m
max_prediction_length = chronos_context_size # 64 intervals
actual_prediction_length = min(prediction_days * 96, max_prediction_length)
trim_length = prediction_days * 96
actual_prediction_length = max(1, actual_prediction_length)
# Use predict_quantiles with proper formatting
with torch.amp.autocast('cuda'):
# Ensure all inputs are on GPU
context = context.to(device)
# Move quantile levels to GPU
quantile_levels = torch.tensor([0.1, 0.5, 0.9], device=device, dtype=dtype)
# Ensure prediction length is on GPU
prediction_length = torch.tensor(actual_prediction_length, device=device, dtype=torch.long)
# Force all model components to GPU
pipe.model = pipe.model.to(device)
# Move model to evaluation mode
pipe.model.eval()
# Ensure context is properly shaped and on GPU
if len(context.shape) == 1:
context = context.unsqueeze(0)
context = context.to(device)
# Move all model parameters and buffers to GPU
for param in pipe.model.parameters():
param.data = param.data.to(device)
for buffer in pipe.model.buffers():
buffer.data = buffer.data.to(device)
# Move all model submodules to GPU
for module in pipe.model.modules():
if hasattr(module, 'to'):
module.to(device)
# Move all model attributes to GPU
for name, value in pipe.model.__dict__.items():
if isinstance(value, torch.Tensor):
pipe.model.__dict__[name] = value.to(device)
# Move all model config tensors to GPU
if hasattr(pipe.model, 'config'):
for key, value in pipe.model.config.__dict__.items():
if isinstance(value, torch.Tensor):
setattr(pipe.model.config, key, value.to(device))
# Move all pipeline tensors to GPU
for name, value in pipe.__dict__.items():
if isinstance(value, torch.Tensor):
setattr(pipe, name, value.to(device))
# Ensure all model states are on GPU
if hasattr(pipe.model, 'state_dict'):
state_dict = pipe.model.state_dict()
for key in state_dict:
if isinstance(state_dict[key], torch.Tensor):
state_dict[key] = state_dict[key].to(device)
pipe.model.load_state_dict(state_dict)
# Move any additional components to GPU
if hasattr(pipe, 'tokenizer'):
# Move tokenizer to GPU if it supports it
if hasattr(pipe.tokenizer, 'to'):
pipe.tokenizer = pipe.tokenizer.to(device)
# Move all tokenizer tensors to GPU
for name, value in pipe.tokenizer.__dict__.items():
if isinstance(value, torch.Tensor):
setattr(pipe.tokenizer, name, value.to(device))
# Handle MeanScaleUniformBins specific attributes
if hasattr(pipe.tokenizer, 'bins'):
if isinstance(pipe.tokenizer.bins, torch.Tensor):
pipe.tokenizer.bins = pipe.tokenizer.bins.to(device)
if hasattr(pipe.tokenizer, 'scale'):
if isinstance(pipe.tokenizer.scale, torch.Tensor):
pipe.tokenizer.scale = pipe.tokenizer.scale.to(device)
if hasattr(pipe.tokenizer, 'mean'):
if isinstance(pipe.tokenizer.mean, torch.Tensor):
pipe.tokenizer.mean = pipe.tokenizer.mean.to(device)
# Move any additional tensors in the tokenizer's attributes to GPU
for name, value in pipe.tokenizer.__dict__.items():
if isinstance(value, torch.Tensor):
pipe.tokenizer.__dict__[name] = value.to(device)
# Remove the EOS token handling since MeanScaleUniformBins doesn't use it
if hasattr(pipe.tokenizer, '_append_eos_token'):
# Create a wrapper that just returns the input tensors
def wrapped_append_eos(token_ids, attention_mask):
return token_ids, attention_mask
pipe.tokenizer._append_eos_token = wrapped_append_eos
# Force synchronization again to ensure all tensors are on GPU
torch.cuda.synchronize()
# Ensure all model components are in eval mode
pipe.model.eval()
# Move any additional tensors in the model's config to GPU
if hasattr(pipe.model, 'config'):
for key, value in pipe.model.config.__dict__.items():
if isinstance(value, torch.Tensor):
setattr(pipe.model.config, key, value.to(device))
# Move any additional tensors in the model's state dict to GPU
if hasattr(pipe.model, 'state_dict'):
state_dict = pipe.model.state_dict()
for key in state_dict:
if isinstance(state_dict[key], torch.Tensor):
state_dict[key] = state_dict[key].to(device)
pipe.model.load_state_dict(state_dict)
# Move any additional tensors in the model's buffers to GPU
for name, buffer in pipe.model.named_buffers():
if buffer is not None:
pipe.model.register_buffer(name, buffer.to(device))
# Move any additional tensors in the model's parameters to GPU
for name, param in pipe.model.named_parameters():
if param is not None:
param.data = param.data.to(device)
# Move any additional tensors in the model's attributes to GPU
for name, value in pipe.model.__dict__.items():
if isinstance(value, torch.Tensor):
pipe.model.__dict__[name] = value.to(device)
# Move any additional tensors in the model's modules to GPU
for name, module in pipe.model.named_modules():
if hasattr(module, 'to'):
module.to(device)
# Move any tensors in the module's __dict__
for key, value in module.__dict__.items():
if isinstance(value, torch.Tensor):
setattr(module, key, value.to(device))
# Force synchronization again to ensure all tensors are on GPU
torch.cuda.synchronize()
# Ensure tokenizer is on GPU and all its tensors are on GPU
if hasattr(pipe, 'tokenizer'):
# Move tokenizer to GPU if it supports it
if hasattr(pipe.tokenizer, 'to'):
pipe.tokenizer = pipe.tokenizer.to(device)
# Move all tokenizer tensors to GPU
for name, value in pipe.tokenizer.__dict__.items():
if isinstance(value, torch.Tensor):
setattr(pipe.tokenizer, name, value.to(device))
# Handle MeanScaleUniformBins specific attributes
if hasattr(pipe.tokenizer, 'bins'):
if isinstance(pipe.tokenizer.bins, torch.Tensor):
pipe.tokenizer.bins = pipe.tokenizer.bins.to(device)
if hasattr(pipe.tokenizer, 'scale'):
if isinstance(pipe.tokenizer.scale, torch.Tensor):
pipe.tokenizer.scale = pipe.tokenizer.scale.to(device)
if hasattr(pipe.tokenizer, 'mean'):
if isinstance(pipe.tokenizer.mean, torch.Tensor):
pipe.tokenizer.mean = pipe.tokenizer.mean.to(device)
# Move any additional tensors in the tokenizer's attributes to GPU
for name, value in pipe.tokenizer.__dict__.items():
if isinstance(value, torch.Tensor):
pipe.tokenizer.__dict__[name] = value.to(device)
# Force synchronization again to ensure all tensors are on GPU
torch.cuda.synchronize()
# Make prediction
quantiles, mean = pipe.predict_quantiles(
context=context,
prediction_length=actual_prediction_length,
quantile_levels=[0.1, 0.5, 0.9]
)
if quantiles is None or mean is None:
raise ValueError("Chronos returned empty prediction")
print(f"Quantiles shape: {quantiles.shape}, Mean shape: {mean.shape}")
# Convert to numpy arrays
quantiles = quantiles.detach().cpu().numpy()
mean = mean.detach().cpu().numpy()
# Denormalize predictions using the same scaler as context
mean_pred = scaler.inverse_transform(mean.reshape(-1, 1)).flatten()
lower_bound = scaler.inverse_transform(quantiles[0, :, 0].reshape(-1, 1)).flatten()
upper_bound = scaler.inverse_transform(quantiles[0, :, 2].reshape(-1, 1)).flatten()
# Calculate standard deviation from quantiles
std_pred = (upper_bound - lower_bound) / (2 * 1.645)
# Check for discontinuity and apply continuity correction
last_actual = prices[-1]
first_pred = mean_pred[0]
if abs(first_pred - last_actual) > max(1e-6, 0.005 * abs(last_actual)): # Further reduced threshold
print(f"Warning: Discontinuity detected between last actual ({last_actual}) and first prediction ({first_pred})")
# Apply continuity correction to first prediction
mean_pred[0] = last_actual
# Adjust subsequent predictions to maintain trend with optional smoothing
if len(mean_pred) > 1:
# Calculate the trend from the original prediction
original_trend = mean_pred[1] - first_pred
# Apply the same trend but starting from the last actual value
for i in range(1, len(mean_pred)):
mean_pred[i] = last_actual + original_trend * i
# Apply financial smoothing if enabled
if use_smoothing:
mean_pred = apply_financial_smoothing(mean_pred, smoothing_type, smoothing_window, smoothing_alpha, 3, use_smoothing)
# If we had to limit the prediction length, extend the prediction recursively
if actual_prediction_length < trim_length:
extended_mean_pred = mean_pred.copy()
extended_std_pred = std_pred.copy()
# Store the original scaler for consistency
original_scaler = scaler
# Calculate the number of extension steps needed
remaining_steps = trim_length - actual_prediction_length
steps_needed = (remaining_steps + actual_prediction_length - 1) // actual_prediction_length
for step in range(steps_needed):
# Use all available datapoints for context, including predictions
# This allows the model to build upon its own predictions for better long-horizon forecasting
all_available_data = np.concatenate([prices, extended_mean_pred])
# If we have more data than chronos_context_size, use the most recent chronos_context_size points
# Otherwise, use all available data (this allows for longer context when available)
if len(all_available_data) > chronos_context_size:
context_window = all_available_data[-chronos_context_size:]
else:
context_window = all_available_data
# Use the original scaler to maintain consistency - fit on historical data only
# but transform the combined context window
normalized_context = original_scaler.transform(context_window.reshape(-1, 1)).flatten()
context = torch.tensor(normalized_context, dtype=dtype, device=device)
if len(context.shape) == 1:
context = context.unsqueeze(0)
# Calculate next prediction length based on timeframe
if timeframe == "1d":
next_length = min(max_prediction_length, remaining_steps)
elif timeframe == "1h":
next_length = min(max_prediction_length, remaining_steps)
else:
next_length = min(max_prediction_length, remaining_steps)
with torch.amp.autocast('cuda'):
next_quantiles, next_mean = pipe.predict_quantiles(
context=context,
prediction_length=next_length,
quantile_levels=[0.1, 0.5, 0.9]
)
# Convert predictions to numpy and denormalize using original scaler
next_mean = next_mean.detach().cpu().numpy()
next_quantiles = next_quantiles.detach().cpu().numpy()
# Denormalize predictions using the original scaler
next_mean_pred = original_scaler.inverse_transform(next_mean.reshape(-1, 1)).flatten()
next_lower = original_scaler.inverse_transform(next_quantiles[0, :, 0].reshape(-1, 1)).flatten()
next_upper = original_scaler.inverse_transform(next_quantiles[0, :, 2].reshape(-1, 1)).flatten()
# Calculate standard deviation
next_std_pred = (next_upper - next_lower) / (2 * 1.645)
# Check for discontinuity and apply continuity correction
if abs(next_mean_pred[0] - extended_mean_pred[-1]) > max(1e-6, 0.05 * abs(extended_mean_pred[-1])):
print(f"Warning: Discontinuity detected between last prediction ({extended_mean_pred[-1]}) and next prediction ({next_mean_pred[0]})")
# Apply continuity correction to first prediction
next_mean_pred[0] = extended_mean_pred[-1]
# Adjust subsequent predictions to maintain trend
if len(next_mean_pred) > 1:
original_trend = next_mean_pred[1] - next_mean_pred[0]
for i in range(1, len(next_mean_pred)):
next_mean_pred[i] = extended_mean_pred[-1] + original_trend * i
# Apply financial smoothing if enabled
if use_smoothing and len(next_mean_pred) > 1:
next_mean_pred = apply_financial_smoothing(next_mean_pred, smoothing_type, smoothing_window, smoothing_alpha, 3, use_smoothing)
# Append predictions
extended_mean_pred = np.concatenate([extended_mean_pred, next_mean_pred])
extended_std_pred = np.concatenate([extended_std_pred, next_std_pred])
remaining_steps -= len(next_mean_pred)
if remaining_steps <= 0:
break
# Trim to exact prediction length if needed
mean_pred = extended_mean_pred[:trim_length]
std_pred = extended_std_pred[:trim_length]
# Extend Chronos forecasting to volume and technical indicators
volume_pred = None
rsi_pred = None
macd_pred = None
try:
# Prepare volume data for Chronos
volume_data = df['Volume'].values
if len(volume_data) >= chronos_context_size:
# Normalize volume data
scaler_range = min(len(volume_data), chronos_context_size * 2)
context_window = volume_data[-chronos_context_size:]
volume_scaler = MinMaxScaler(feature_range=(-1, 1))
# Fit scaler on a larger range for better normalization
volume_scaler.fit(volume_data[-scaler_range:].reshape(-1, 1))
normalized_volume = volume_scaler.transform(context_window.reshape(-1, 1)).flatten()
if len(normalized_volume) < chronos_context_size:
padding = np.full(chronos_context_size - len(normalized_volume), normalized_volume[-1])
normalized_volume = np.concatenate([padding, normalized_volume])
elif len(normalized_volume) > chronos_context_size:
normalized_volume = normalized_volume[-chronos_context_size:]
volume_context = torch.tensor(normalized_volume, dtype=dtype, device=device)
if len(volume_context.shape) == 1:
volume_context = volume_context.unsqueeze(0)
with torch.amp.autocast('cuda'):
volume_quantiles, volume_mean = pipe.predict_quantiles(
context=volume_context,
prediction_length=actual_prediction_length,
quantile_levels=[0.1, 0.5, 0.9]
)
volume_quantiles = volume_quantiles.detach().cpu().numpy()
volume_mean = volume_mean.detach().cpu().numpy()
volume_pred = volume_scaler.inverse_transform(volume_mean.reshape(-1, 1)).flatten()
lower_bound = volume_scaler.inverse_transform(volume_quantiles[0, :, 0].reshape(-1, 1)).flatten()
upper_bound = volume_scaler.inverse_transform(volume_quantiles[0, :, 2].reshape(-1, 1)).flatten()
std_pred_vol = (upper_bound - lower_bound) / (2 * 1.645)
last_actual = volume_data[-1]
first_pred = volume_pred[0]
if abs(first_pred - last_actual) > max(1e-6, 0.005 * abs(last_actual)):
print(f"Warning: Discontinuity detected between last actual volume ({last_actual}) and first prediction ({first_pred})")
# Apply continuity correction
volume_pred[0] = last_actual
# Adjust subsequent predictions to maintain trend with optional smoothing
if len(volume_pred) > 1:
# Calculate the trend from the original prediction
original_trend = volume_pred[1] - first_pred
# Apply the same trend but starting from the last actual value
for i in range(1, len(volume_pred)):
volume_pred[i] = last_actual + original_trend * i
# Apply financial smoothing if enabled
if use_smoothing:
volume_pred = apply_financial_smoothing(volume_pred, smoothing_type, smoothing_window, smoothing_alpha, 3, use_smoothing)
# Extend volume predictions if needed
if actual_prediction_length < trim_length:
extended_volume_pred = volume_pred.copy()
extended_volume_std = std_pred_vol.copy()
remaining_steps = trim_length - actual_prediction_length
steps_needed = (remaining_steps + actual_prediction_length - 1) // actual_prediction_length
for step in range(steps_needed):
# Use all available datapoints for context, including predictions
# This allows the model to build upon its own predictions for better long-horizon forecasting
all_available_data = np.concatenate([volume_data, extended_volume_pred])
# If we have more data than chronos_context_size, use the most recent chronos_context_size points
# Otherwise, use all available data (this allows for longer context when available)
if len(all_available_data) > chronos_context_size:
context_window = all_available_data[-chronos_context_size:]
else:
context_window = all_available_data
# Use the original volume scaler to maintain consistency - fit on historical data only
# but transform the combined context window
normalized_context = volume_scaler.transform(context_window.reshape(-1, 1)).flatten()
context = torch.tensor(normalized_context, dtype=dtype, device=device)
if len(context.shape) == 1:
context = context.unsqueeze(0)
next_length = min(chronos_context_size, remaining_steps)
with torch.amp.autocast('cuda'):
next_quantiles, next_mean = pipe.predict_quantiles(
context=context,
prediction_length=next_length,
quantile_levels=[0.1, 0.5, 0.9]
)
next_mean = next_mean.detach().cpu().numpy()
next_quantiles = next_quantiles.detach().cpu().numpy()
next_mean_pred = volume_scaler.inverse_transform(next_mean.reshape(-1, 1)).flatten()
next_lower = volume_scaler.inverse_transform(next_quantiles[0, :, 0].reshape(-1, 1)).flatten()
next_upper = volume_scaler.inverse_transform(next_quantiles[0, :, 2].reshape(-1, 1)).flatten()
next_std_pred = (next_upper - next_lower) / (2 * 1.645)
# Check for discontinuity and apply continuity correction
if abs(next_mean_pred[0] - extended_volume_pred[-1]) > max(1e-6, 0.05 * abs(extended_volume_pred[-1])):
print(f"Warning: Discontinuity detected between last volume prediction ({extended_volume_pred[-1]}) and next prediction ({next_mean_pred[0]})")
next_mean_pred[0] = extended_volume_pred[-1]
if len(next_mean_pred) > 1:
original_trend = next_mean_pred[1] - next_mean_pred[0]
for i in range(1, len(next_mean_pred)):
next_mean_pred[i] = extended_volume_pred[-1] + original_trend * i
# Apply financial smoothing if enabled
if use_smoothing and len(next_mean_pred) > 1:
next_mean_pred = apply_financial_smoothing(next_mean_pred, smoothing_type, smoothing_window, smoothing_alpha, 3, use_smoothing)
extended_volume_pred = np.concatenate([extended_volume_pred, next_mean_pred])
extended_volume_std = np.concatenate([extended_volume_std, next_std_pred])
remaining_steps -= len(next_mean_pred)
if remaining_steps <= 0:
break
volume_pred = extended_volume_pred[:trim_length]
else:
avg_volume = df['Volume'].mean()
volume_pred = np.full(trim_length, avg_volume)
except Exception as e:
print(f"Volume prediction error: {str(e)}")
# Fallback: use historical average
avg_volume = df['Volume'].mean()
volume_pred = np.full(trim_length, avg_volume)
try:
# Prepare RSI data for Chronos
rsi_data = df['RSI'].values
if len(rsi_data) >= chronos_context_size and not np.any(np.isnan(rsi_data)):
# RSI is already normalized (0-100), but we'll scale it to (-1, 1)
scaler_range = min(len(rsi_data), chronos_context_size * 2)
context_window = rsi_data[-chronos_context_size:]
rsi_scaler = MinMaxScaler(feature_range=(-1, 1))
# Fit scaler on a larger range for better normalization
rsi_scaler.fit(rsi_data[-scaler_range:].reshape(-1, 1))
normalized_rsi = rsi_scaler.transform(context_window.reshape(-1, 1)).flatten()
if len(normalized_rsi) < chronos_context_size:
padding = np.full(chronos_context_size - len(normalized_rsi), normalized_rsi[-1])
normalized_rsi = np.concatenate([padding, normalized_rsi])
elif len(normalized_rsi) > chronos_context_size:
normalized_rsi = normalized_rsi[-chronos_context_size:]
rsi_context = torch.tensor(normalized_rsi, dtype=dtype, device=device)
if len(rsi_context.shape) == 1:
rsi_context = rsi_context.unsqueeze(0)
with torch.amp.autocast('cuda'):
rsi_quantiles, rsi_mean = pipe.predict_quantiles(
context=rsi_context,
prediction_length=actual_prediction_length,
quantile_levels=[0.1, 0.5, 0.9]
)
# Convert and denormalize RSI predictions
rsi_quantiles = rsi_quantiles.detach().cpu().numpy()
rsi_mean = rsi_mean.detach().cpu().numpy()
rsi_pred = rsi_scaler.inverse_transform(rsi_mean.reshape(-1, 1)).flatten()
# Clamp RSI to valid range (0-100)
lower_bound = rsi_scaler.inverse_transform(rsi_quantiles[0, :, 0].reshape(-1, 1)).flatten()
upper_bound = rsi_scaler.inverse_transform(rsi_quantiles[0, :, 2].reshape(-1, 1)).flatten()
std_pred_rsi = (upper_bound - lower_bound) / (2 * 1.645)
rsi_pred = np.clip(rsi_pred, 0, 100)
last_actual = rsi_data[-1]
first_pred = rsi_pred[0]
if abs(first_pred - last_actual) > max(1e-6, 0.005 * abs(last_actual)):
print(f"Warning: Discontinuity detected between last actual RSI ({last_actual}) and first prediction ({first_pred})")
# Apply continuity correction
rsi_pred[0] = last_actual
if len(rsi_pred) > 1:
trend = rsi_pred[1] - first_pred
rsi_pred[1:] = rsi_pred[1:] - first_pred + last_actual
rsi_pred = np.clip(rsi_pred, 0, 100) # Re-clip after adjustment
# Extend RSI predictions if needed
if actual_prediction_length < trim_length:
extended_rsi_pred = rsi_pred.copy()
extended_rsi_std = std_pred_rsi.copy()
remaining_steps = trim_length - actual_prediction_length
steps_needed = (remaining_steps + actual_prediction_length - 1) // actual_prediction_length
for step in range(steps_needed):
# Use all available datapoints for context, including predictions
# This allows the model to build upon its own predictions for better long-horizon forecasting
all_available_data = np.concatenate([rsi_data, extended_rsi_pred])
# If we have more data than chronos_context_size, use the most recent chronos_context_size points
# Otherwise, use all available data (this allows for longer context when available)
if len(all_available_data) > chronos_context_size:
context_window = all_available_data[-chronos_context_size:]
else:
context_window = all_available_data
# Use the original RSI scaler to maintain consistency - fit on historical data only
# but transform the combined context window
normalized_context = rsi_scaler.transform(context_window.reshape(-1, 1)).flatten()
context = torch.tensor(normalized_context, dtype=dtype, device=device)
if len(context.shape) == 1:
context = context.unsqueeze(0)
next_length = min(chronos_context_size, remaining_steps)
with torch.amp.autocast('cuda'):
next_quantiles, next_mean = pipe.predict_quantiles(
context=context,
prediction_length=next_length,
quantile_levels=[0.1, 0.5, 0.9]
)
next_mean = next_mean.detach().cpu().numpy()
next_quantiles = next_quantiles.detach().cpu().numpy()
next_mean_pred = rsi_scaler.inverse_transform(next_mean.reshape(-1, 1)).flatten()
next_lower = rsi_scaler.inverse_transform(next_quantiles[0, :, 0].reshape(-1, 1)).flatten()
next_upper = rsi_scaler.inverse_transform(next_quantiles[0, :, 2].reshape(-1, 1)).flatten()
next_std_pred = (next_upper - next_lower) / (2 * 1.645)
next_mean_pred = np.clip(next_mean_pred, 0, 100)
# Check for discontinuity and apply continuity correction
if abs(next_mean_pred[0] - extended_rsi_pred[-1]) > max(1e-6, 0.005 * abs(extended_rsi_pred[-1])):
print(f"Warning: Discontinuity detected between last RSI prediction ({extended_rsi_pred[-1]}) and next prediction ({next_mean_pred[0]})")
next_mean_pred[0] = extended_rsi_pred[-1]
if len(next_mean_pred) > 1:
original_trend = next_mean_pred[1] - next_mean_pred[0]
for i in range(1, len(next_mean_pred)):
next_mean_pred[i] = extended_rsi_pred[-1] + original_trend * i
next_mean_pred = np.clip(next_mean_pred, 0, 100)
# Apply financial smoothing if enabled
if use_smoothing and len(next_mean_pred) > 1:
next_mean_pred = apply_financial_smoothing(next_mean_pred, smoothing_type, smoothing_window, smoothing_alpha, 3, use_smoothing)
next_mean_pred = np.clip(next_mean_pred, 0, 100)
extended_rsi_pred = np.concatenate([extended_rsi_pred, next_mean_pred])
extended_rsi_std = np.concatenate([extended_rsi_std, next_std_pred])
remaining_steps -= len(next_mean_pred)
if remaining_steps <= 0:
break
rsi_pred = extended_rsi_pred[:trim_length]
else:
last_rsi = df['RSI'].iloc[-1]
rsi_pred = np.full(trim_length, last_rsi)
except Exception as e:
print(f"RSI prediction error: {str(e)}")
# Fallback: use last known RSI value
last_rsi = df['RSI'].iloc[-1]
rsi_pred = np.full(trim_length, last_rsi)
try:
# Prepare MACD data for Chronos
macd_data = df['MACD'].values
if len(macd_data) >= chronos_context_size and not np.any(np.isnan(macd_data)):
# Normalize MACD data
scaler_range = min(len(macd_data), chronos_context_size * 2)
context_window = macd_data[-chronos_context_size:]
macd_scaler = MinMaxScaler(feature_range=(-1, 1))
# Fit scaler on a larger range for better normalization
macd_scaler.fit(macd_data[-scaler_range:].reshape(-1, 1))
normalized_macd = macd_scaler.transform(context_window.reshape(-1, 1)).flatten()
if len(normalized_macd) < chronos_context_size:
padding = np.full(chronos_context_size - len(normalized_macd), normalized_macd[-1])
normalized_macd = np.concatenate([padding, normalized_macd])
elif len(normalized_macd) > chronos_context_size:
normalized_macd = normalized_macd[-chronos_context_size:]
macd_context = torch.tensor(normalized_macd, dtype=dtype, device=device)
if len(macd_context.shape) == 1:
macd_context = macd_context.unsqueeze(0)
with torch.amp.autocast('cuda'):
macd_quantiles, macd_mean = pipe.predict_quantiles(
context=macd_context,
prediction_length=actual_prediction_length,
quantile_levels=[0.1, 0.5, 0.9]
)
# Convert and denormalize MACD predictions
macd_quantiles = macd_quantiles.detach().cpu().numpy()
macd_mean = macd_mean.detach().cpu().numpy()
macd_pred = macd_scaler.inverse_transform(macd_mean.reshape(-1, 1)).flatten()
lower_bound = macd_scaler.inverse_transform(macd_quantiles[0, :, 0].reshape(-1, 1)).flatten()
upper_bound = macd_scaler.inverse_transform(macd_quantiles[0, :, 2].reshape(-1, 1)).flatten()
std_pred_macd = (upper_bound - lower_bound) / (2 * 1.645)
last_actual = macd_data[-1]
first_pred = macd_pred[0]
# Check for discontinuity and apply continuity correction
if abs(first_pred - last_actual) > max(1e-6, 0.005 * abs(last_actual)):
print(f"Warning: Discontinuity detected between last actual MACD ({last_actual}) and first prediction ({first_pred})")
# Apply continuity correction
macd_pred[0] = last_actual
# Adjust subsequent predictions to maintain trend with optional smoothing
if len(macd_pred) > 1:
# Calculate the trend from the original prediction
original_trend = macd_pred[1] - first_pred
# Apply the same trend but starting from the last actual value
for i in range(1, len(macd_pred)):
macd_pred[i] = last_actual + original_trend * i
# Apply financial smoothing if enabled
if use_smoothing:
macd_pred = apply_financial_smoothing(macd_pred, smoothing_type, smoothing_window, smoothing_alpha, 3, use_smoothing)
# Extend MACD predictions if needed
if actual_prediction_length < trim_length:
extended_macd_pred = macd_pred.copy()
extended_macd_std = std_pred_macd.copy()
remaining_steps = trim_length - actual_prediction_length
steps_needed = (remaining_steps + actual_prediction_length - 1) // actual_prediction_length
for step in range(steps_needed):
# Use all available datapoints for context, including predictions
# This allows the model to build upon its own predictions for better long-horizon forecasting
all_available_data = np.concatenate([macd_data, extended_macd_pred])
# If we have more data than chronos_context_size, use the most recent chronos_context_size points
# Otherwise, use all available data (this allows for longer context when available)
if len(all_available_data) > chronos_context_size:
context_window = all_available_data[-chronos_context_size:]
else:
context_window = all_available_data
# Use the original MACD scaler to maintain consistency - fit on historical data only
# but transform the combined context window
normalized_context = macd_scaler.transform(context_window.reshape(-1, 1)).flatten()
context = torch.tensor(normalized_context, dtype=dtype, device=device)
if len(context.shape) == 1:
context = context.unsqueeze(0)
next_length = min(chronos_context_size, remaining_steps)
with torch.amp.autocast('cuda'):
next_quantiles, next_mean = pipe.predict_quantiles(
context=context,
prediction_length=next_length,
quantile_levels=[0.1, 0.5, 0.9]
)
next_mean = next_mean.detach().cpu().numpy()
next_quantiles = next_quantiles.detach().cpu().numpy()
next_mean_pred = macd_scaler.inverse_transform(next_mean.reshape(-1, 1)).flatten()
next_lower = macd_scaler.inverse_transform(next_quantiles[0, :, 0].reshape(-1, 1)).flatten()
next_upper = macd_scaler.inverse_transform(next_quantiles[0, :, 2].reshape(-1, 1)).flatten()
next_std_pred = (next_upper - next_lower) / (2 * 1.645)
# Check for discontinuity and apply continuity correction
if abs(next_mean_pred[0] - extended_macd_pred[-1]) > max(1e-6, 0.05 * abs(extended_macd_pred[-1])):
print(f"Warning: Discontinuity detected between last MACD prediction ({extended_macd_pred[-1]}) and next prediction ({next_mean_pred[0]})")
next_mean_pred[0] = extended_macd_pred[-1]
if len(next_mean_pred) > 1:
original_trend = next_mean_pred[1] - next_mean_pred[0]
for i in range(1, len(next_mean_pred)):
next_mean_pred[i] = extended_macd_pred[-1] + original_trend * i
# Apply financial smoothing if enabled
if use_smoothing and len(next_mean_pred) > 1:
next_mean_pred = apply_financial_smoothing(next_mean_pred, smoothing_type, smoothing_window, smoothing_alpha, 3, use_smoothing)
extended_macd_pred = np.concatenate([extended_macd_pred, next_mean_pred])
extended_macd_std = np.concatenate([extended_macd_std, next_std_pred])
remaining_steps -= len(next_mean_pred)
if remaining_steps <= 0:
break
macd_pred = extended_macd_pred[:trim_length]
else:
last_macd = df['MACD'].iloc[-1]
macd_pred = np.full(trim_length, last_macd)
except Exception as e:
print(f"MACD prediction error: {str(e)}")
# Fallback: use last known MACD value
last_macd = df['MACD'].iloc[-1]
macd_pred = np.full(trim_length, last_macd)
except Exception as e:
print(f"Chronos prediction error: {str(e)}")
print(f"Error type: {type(e)}")
print(f"Error details: {str(e)}")
raise
if strategy == "technical":
# Technical analysis based prediction
last_price = df['Close'].iloc[-1]
rsi = df['RSI'].iloc[-1]
macd = df['MACD'].iloc[-1]
macd_signal = df['MACD_Signal'].iloc[-1]
# Simple prediction based on technical indicators
trend = 1 if (rsi > 50 and macd > macd_signal) else -1
volatility = df['Volatility'].iloc[-1]
# Generate predictions
mean_pred = np.array([last_price * (1 + trend * volatility * i) for i in range(1, prediction_days + 1)])
std_pred = np.array([volatility * last_price * i for i in range(1, prediction_days + 1)])
# Create prediction dates based on timeframe
last_date = df.index[-1]
if timeframe == "1d":
pred_dates = pd.date_range(start=last_date + timedelta(days=1), periods=prediction_days)
elif timeframe == "1h":
pred_dates = pd.date_range(start=last_date + timedelta(hours=1), periods=prediction_days * 24)
else: # 15m
pred_dates = pd.date_range(start=last_date + timedelta(minutes=15), periods=prediction_days * 96)
# Create visualization
fig = make_subplots(rows=3, cols=1,
shared_xaxes=True,
vertical_spacing=0.05,
subplot_titles=('Price Prediction', 'Technical Indicators', 'Volume'))
# Add historical price
fig.add_trace(
go.Scatter(x=df.index, y=df['Close'], name='Historical Price',
line=dict(color='blue')),
row=1, col=1
)
# Add prediction mean
fig.add_trace(
go.Scatter(x=pred_dates, y=mean_pred, name='Predicted Price',
line=dict(color='red')),
row=1, col=1
)
# Add confidence intervals
fig.add_trace(
go.Scatter(x=pred_dates, y=mean_pred + 1.96 * std_pred,
fill=None, mode='lines', line_color='rgba(255,0,0,0.2)',
name='Upper Bound'),
row=1, col=1
)
fig.add_trace(
go.Scatter(x=pred_dates, y=mean_pred - 1.96 * std_pred,
fill='tonexty', mode='lines', line_color='rgba(255,0,0,0.2)',
name='Lower Bound'),
row=1, col=1
)
# Add technical indicators
fig.add_trace(
go.Scatter(x=df.index, y=df['RSI'], name='RSI',
line=dict(color='purple')),
row=2, col=1
)
fig.add_trace(
go.Scatter(x=df.index, y=df['MACD'], name='MACD',
line=dict(color='orange')),
row=2, col=1
)
fig.add_trace(
go.Scatter(x=df.index, y=df['MACD_Signal'], name='MACD Signal',
line=dict(color='green')),
row=2, col=1
)
# Add predicted technical indicators if available
if rsi_pred is not None:
fig.add_trace(
go.Scatter(x=pred_dates, y=rsi_pred, name='Predicted RSI',
line=dict(color='purple', dash='dash')),
row=2, col=1
)
if macd_pred is not None:
fig.add_trace(
go.Scatter(x=pred_dates, y=macd_pred, name='Predicted MACD',
line=dict(color='orange', dash='dash')),
row=2, col=1
)
# Add volume
fig.add_trace(
go.Bar(x=df.index, y=df['Volume'], name='Volume',
marker_color='gray'),
row=3, col=1
)
# Add predicted volume if available
if volume_pred is not None:
fig.add_trace(
go.Bar(x=pred_dates, y=volume_pred, name='Predicted Volume',
marker_color='red', opacity=0.7),
row=3, col=1
)
# Update layout with timeframe-specific settings
fig.update_layout(
title=f'{symbol} {timeframe} Analysis and Prediction',
xaxis_title='Date',
yaxis_title='Price',
height=1000,
showlegend=True
)
# Calculate trading signals
signals = calculate_trading_signals(df)
# Add prediction information to signals
signals.update({
"symbol": symbol,
"timeframe": timeframe,
"prediction": mean_pred.tolist(),
"confidence": std_pred.tolist(),
"dates": pred_dates.strftime('%Y-%m-%d %H:%M:%S').tolist(),
"strategy_used": strategy
})
# Add predicted indicators to signals if available
if volume_pred is not None:
signals["predicted_volume"] = volume_pred.tolist()
if rsi_pred is not None:
signals["predicted_rsi"] = rsi_pred.tolist()
if macd_pred is not None:
signals["predicted_macd"] = macd_pred.tolist()
# Implement advanced features
# 1. Market Regime Detection
if use_regime_detection:
try:
returns = df['Returns'].dropna()
regime_info = detect_market_regime(returns)
signals["regime_info"] = regime_info
except Exception as e:
print(f"Regime detection error: {str(e)}")
signals["regime_info"] = {"error": str(e)}
# 2. Advanced Trading Signals with Regime Awareness
try:
regime_info = signals.get("regime_info", {})
advanced_signals = advanced_trading_signals(df, regime_info)
signals["advanced_signals"] = advanced_signals
except Exception as e:
print(f"Advanced trading signals error: {str(e)}")
signals["advanced_signals"] = {"error": str(e)}
# 3. Stress Testing
if use_stress_testing:
try:
stress_results = stress_test_scenarios(df, mean_pred)
signals["stress_test_results"] = stress_results
except Exception as e:
print(f"Stress testing error: {str(e)}")
signals["stress_test_results"] = {"error": str(e)}
# 4. Ensemble Methods
if use_ensemble and ensemble_weights:
try:
ensemble_mean, ensemble_uncertainty = create_ensemble_prediction(
df, prediction_days, ensemble_weights
)
if len(ensemble_mean) > 0:
signals["ensemble_used"] = True
signals["ensemble_prediction"] = ensemble_mean.tolist()
signals["ensemble_uncertainty"] = ensemble_uncertainty.tolist()
# Update the main prediction with ensemble if available
if len(ensemble_mean) == len(mean_pred):
mean_pred = ensemble_mean
std_pred = ensemble_uncertainty
else:
signals["ensemble_used"] = False
except Exception as e:
print(f"Ensemble prediction error: {str(e)}")
signals["ensemble_used"] = False
signals["ensemble_error"] = str(e)
# 5. Enhanced Uncertainty Quantification
try:
if 'quantiles' in locals():
skewed_uncertainty = calculate_skewed_uncertainty(quantiles)
signals["skewed_uncertainty"] = skewed_uncertainty.tolist()
except Exception as e:
print(f"Skewed uncertainty calculation error: {str(e)}")
return signals, fig
except Exception as e:
raise Exception(f"Prediction error: {str(e)}")
finally:
clear_gpu_memory()
def calculate_trading_signals(df: pd.DataFrame) -> Dict:
"""Calculate trading signals based on technical indicators"""
signals = {
"RSI": "Oversold" if df['RSI'].iloc[-1] < 30 else "Overbought" if df['RSI'].iloc[-1] > 70 else "Neutral",
"MACD": "Buy" if df['MACD'].iloc[-1] > df['MACD_Signal'].iloc[-1] else "Sell",
"Bollinger": "Buy" if df['Close'].iloc[-1] < df['BB_Lower'].iloc[-1] else "Sell" if df['Close'].iloc[-1] > df['BB_Upper'].iloc[-1] else "Hold",
"SMA": "Buy" if df['SMA_20'].iloc[-1] > df['SMA_50'].iloc[-1] else "Sell"
}
# Calculate overall signal
buy_signals = sum(1 for signal in signals.values() if signal == "Buy")
sell_signals = sum(1 for signal in signals.values() if signal == "Sell")
if buy_signals > sell_signals:
signals["Overall"] = "Buy"
elif sell_signals > buy_signals:
signals["Overall"] = "Sell"
else:
signals["Overall"] = "Hold"
return signals
def get_market_data(symbol: str = "^GSPC", lookback_days: int = 365) -> pd.DataFrame:
"""
Fetch market data (S&P 500 by default) for correlation analysis and regime detection.
Args:
symbol (str): Market index symbol (default: ^GSPC for S&P 500)
lookback_days (int): Number of days to look back
Returns:
pd.DataFrame: Market data with returns
"""
cache_key = f"{symbol}_{lookback_days}"
current_time = time.time()
# Check cache
if cache_key in market_data_cache and current_time < cache_expiry.get(cache_key, 0):
return market_data_cache[cache_key]
try:
ticker = yf.Ticker(symbol)
end_date = datetime.now()
start_date = end_date - timedelta(days=lookback_days)
def fetch_market_history():
return ticker.history(
start=start_date,
end=end_date,
interval="1d",
prepost=False,
actions=False,
auto_adjust=True
)
df = retry_yfinance_request(fetch_market_history)
if not df.empty:
df['Returns'] = df['Close'].pct_change()
df['Volatility'] = df['Returns'].rolling(window=20).std()
# Cache the data
market_data_cache[cache_key] = df
cache_expiry[cache_key] = current_time + CACHE_DURATION
return df
except Exception as e:
print(f"Warning: Could not fetch market data for {symbol}: {str(e)}")
return pd.DataFrame()
def detect_market_regime(returns: pd.Series, n_regimes: int = 3) -> Dict:
"""
Detect market regime using Hidden Markov Model or simplified methods.
Args:
returns (pd.Series): Price returns
n_regimes (int): Number of regimes to detect
Returns:
Dict: Regime information including probabilities and characteristics
"""
def get_regime_name(regime_idx: int, means: List[float], volatilities: List[float]) -> str:
"""
Convert regime index to descriptive name based on characteristics.
Args:
regime_idx (int): Regime index (0, 1, 2)
means (List[float]): List of regime means
volatilities (List[float]): List of regime volatilities
Returns:
str: Descriptive regime name
"""
if len(means) != 3 or len(volatilities) != 3:
return f"Regime {regime_idx}"
# Sort regimes by volatility (low to high)
vol_sorted = sorted(range(len(volatilities)), key=lambda i: volatilities[i])
# Sort regimes by mean return (low to high)
mean_sorted = sorted(range(len(means)), key=lambda i: means[i])
# Determine regime characteristics
if regime_idx == vol_sorted[0]: # Lowest volatility
if means[regime_idx] > 0:
return "Low Volatility Bull"
else:
return "Low Volatility Bear"
elif regime_idx == vol_sorted[2]: # Highest volatility
if means[regime_idx] > 0:
return "High Volatility Bull"
else:
return "High Volatility Bear"
else: # Medium volatility
if means[regime_idx] > 0:
return "Moderate Bull"
else:
return "Moderate Bear"
if len(returns) < 50:
return {"regime": "Normal Market", "probabilities": [1.0], "volatility": returns.std()}
try:
if HMM_AVAILABLE:
# Use HMM for regime detection
# Convert pandas Series to numpy array for reshape
returns_array = returns.dropna().values
# Try different HMM configurations if convergence fails
for attempt in range(3):
try:
if attempt == 0:
model = hmm.GaussianHMM(n_components=n_regimes, random_state=42, covariance_type="full", n_iter=100)
elif attempt == 1:
model = hmm.GaussianHMM(n_components=n_regimes, random_state=42, covariance_type="diag", n_iter=200)
else:
model = hmm.GaussianHMM(n_components=n_regimes, random_state=42, covariance_type="spherical", n_iter=300)
model.fit(returns_array.reshape(-1, 1))
# Get regime probabilities for the last observation
regime_probs = model.predict_proba(returns_array.reshape(-1, 1))
current_regime = model.predict(returns_array.reshape(-1, 1))[-1]
# Calculate regime characteristics
regime_means = model.means_.flatten()
regime_vols = np.sqrt(model.covars_.diagonal(axis1=1, axis2=2)) if model.covariance_type == "full" else np.sqrt(model.covars_)
# Convert regime index to descriptive name
regime_name = get_regime_name(int(current_regime), regime_means.tolist(), regime_vols.tolist())
return {
"regime": regime_name,
"regime_index": int(current_regime),
"probabilities": regime_probs[-1].tolist(),
"means": regime_means.tolist(),
"volatilities": regime_vols.tolist(),
"method": f"HMM-{model.covariance_type}"
}
except Exception as e:
if attempt == 2: # Last attempt failed
print(f"HMM failed after {attempt + 1} attempts: {str(e)}")
break
continue
else:
# Simplified regime detection using volatility clustering
volatility = returns.rolling(window=20).std().dropna()
vol_percentile = volatility.iloc[-1] / volatility.quantile(0.8)
if vol_percentile > 1.2:
regime_name = "High Volatility Market"
regime = 2 # High volatility regime
elif vol_percentile < 0.8:
regime_name = "Low Volatility Market"
regime = 0 # Low volatility regime
else:
regime_name = "Normal Market"
regime = 1 # Normal regime
return {
"regime": regime_name,
"regime_index": regime,
"probabilities": [0.1, 0.8, 0.1] if regime == 1 else [0.8, 0.1, 0.1] if regime == 0 else [0.1, 0.1, 0.8],
"volatility": volatility.iloc[-1],
"method": "Volatility-based"
}
except Exception as e:
print(f"Warning: Regime detection failed: {str(e)}")
return {"regime": "Normal Market", "regime_index": 1, "probabilities": [1.0], "volatility": returns.std(), "method": "Fallback"}
def calculate_advanced_risk_metrics(df: pd.DataFrame, market_returns: pd.Series = None,
risk_free_rate: float = 0.02) -> Dict:
"""
Calculate advanced risk metrics including tail risk and market correlation.
Args:
df (pd.DataFrame): Stock data
market_returns (pd.Series): Market returns for correlation analysis
risk_free_rate (float): Annual risk-free rate
Returns:
Dict: Advanced risk metrics
"""
try:
returns = df['Returns'].dropna()
if len(returns) < 30:
return {"error": "Insufficient data for risk calculation"}
# Basic metrics
annual_return = returns.mean() * 252
annual_vol = returns.std() * np.sqrt(252)
# Market-adjusted metrics
beta = 1.0
alpha = 0.0
correlation = 0.0
aligned_returns = None
aligned_market = None
if market_returns is not None and len(market_returns) > 0:
try:
# Align dates
aligned_returns = returns.reindex(market_returns.index).dropna()
aligned_market = market_returns.reindex(aligned_returns.index).dropna()
# Ensure both arrays have the same length
if len(aligned_returns) > 10 and len(aligned_market) > 10:
# Find the common length
min_length = min(len(aligned_returns), len(aligned_market))
aligned_returns = aligned_returns.iloc[-min_length:]
aligned_market = aligned_market.iloc[-min_length:]
# Ensure they have the same length
if len(aligned_returns) == len(aligned_market) and len(aligned_returns) > 10:
try:
beta = np.cov(aligned_returns, aligned_market)[0,1] / np.var(aligned_market)
alpha = aligned_returns.mean() - beta * aligned_market.mean()
correlation = np.corrcoef(aligned_returns, aligned_market)[0,1]
except Exception as e:
print(f"Market correlation calculation error: {str(e)}")
beta = 1.0
alpha = 0.0
correlation = 0.0
else:
beta = 1.0
alpha = 0.0
correlation = 0.0
else:
beta = 1.0
alpha = 0.0
correlation = 0.0
except Exception as e:
print(f"Market data alignment error: {str(e)}")
beta = 1.0
alpha = 0.0
correlation = 0.0
aligned_returns = None
aligned_market = None
# Tail risk metrics
var_95 = np.percentile(returns, 5)
var_99 = np.percentile(returns, 1)
cvar_95 = returns[returns <= var_95].mean()
cvar_99 = returns[returns <= var_99].mean()
# Maximum drawdown
cumulative_returns = (1 + returns).cumprod()
rolling_max = cumulative_returns.expanding().max()
drawdown = (cumulative_returns - rolling_max) / rolling_max
max_drawdown = drawdown.min()
# Skewness and kurtosis
skewness = stats.skew(returns)
kurtosis = stats.kurtosis(returns)
# Risk-adjusted returns
sharpe_ratio = (annual_return - risk_free_rate) / annual_vol if annual_vol > 0 else 0
sortino_ratio = (annual_return - risk_free_rate) / (returns[returns < 0].std() * np.sqrt(252)) if returns[returns < 0].std() > 0 else 0
calmar_ratio = annual_return / abs(max_drawdown) if max_drawdown != 0 else 0
# Information ratio (if market data available)
information_ratio = 0
if aligned_returns is not None and aligned_market is not None:
try:
if len(aligned_returns) > 10 and len(aligned_market) > 10:
min_length = min(len(aligned_returns), len(aligned_market))
aligned_returns_for_ir = aligned_returns.iloc[-min_length:]
aligned_market_for_ir = aligned_market.iloc[-min_length:]
if len(aligned_returns_for_ir) == len(aligned_market_for_ir):
excess_returns = aligned_returns_for_ir - aligned_market_for_ir
information_ratio = excess_returns.mean() / excess_returns.std() if excess_returns.std() > 0 else 0
else:
information_ratio = 0
else:
information_ratio = 0
except Exception as e:
print(f"Information ratio calculation error: {str(e)}")
information_ratio = 0
return {
"Annual_Return": annual_return,
"Annual_Volatility": annual_vol,
"Sharpe_Ratio": sharpe_ratio,
"Sortino_Ratio": sortino_ratio,
"Calmar_Ratio": calmar_ratio,
"Information_Ratio": information_ratio,
"Beta": beta,
"Alpha": alpha * 252,
"Correlation_with_Market": correlation,
"VaR_95": var_95,
"VaR_99": var_99,
"CVaR_95": cvar_95,
"CVaR_99": cvar_99,
"Max_Drawdown": max_drawdown,
"Skewness": skewness,
"Kurtosis": kurtosis,
"Risk_Free_Rate": risk_free_rate
}
except Exception as e:
print(f"Advanced risk metrics calculation error: {str(e)}")
return {"error": f"Risk calculation failed: {str(e)}"}
def create_ensemble_prediction(df: pd.DataFrame, prediction_days: int,
ensemble_weights: Dict = None) -> Tuple[np.ndarray, np.ndarray]:
"""
Create ensemble prediction combining multiple models.
Args:
df (pd.DataFrame): Historical data
prediction_days (int): Number of days to predict
ensemble_weights (Dict): Weights for different models
Returns:
Tuple[np.ndarray, np.ndarray]: Mean and uncertainty predictions
"""
if ensemble_weights is None:
ensemble_weights = {"chronos": 0.6, "technical": 0.2, "statistical": 0.2}
predictions = {}
uncertainties = {}
# Chronos prediction (placeholder - will be filled by main prediction function)
predictions["chronos"] = np.array([])
uncertainties["chronos"] = np.array([])
# Technical prediction
if ensemble_weights.get("technical", 0) > 0:
try:
last_price = df['Close'].iloc[-1]
rsi = df['RSI'].iloc[-1]
macd = df['MACD'].iloc[-1]
macd_signal = df['MACD_Signal'].iloc[-1]
volatility = df['Volatility'].iloc[-1]
# Enhanced technical prediction
trend = 1 if (rsi > 50 and macd > macd_signal) else -1
mean_reversion = (df['SMA_200'].iloc[-1] - last_price) / last_price if 'SMA_200' in df.columns else 0
tech_pred = []
for i in range(1, prediction_days + 1):
# Combine trend and mean reversion
prediction = last_price * (1 + trend * volatility * 0.3 + mean_reversion * 0.1 * i)
tech_pred.append(prediction)
predictions["technical"] = np.array(tech_pred)
uncertainties["technical"] = np.array([volatility * last_price * i for i in range(1, prediction_days + 1)])
except Exception as e:
print(f"Technical prediction error: {str(e)}")
predictions["technical"] = np.array([])
uncertainties["technical"] = np.array([])
# Statistical prediction (ARIMA-like)
if ensemble_weights.get("statistical", 0) > 0:
try:
returns = df['Returns'].dropna()
if len(returns) > 10:
# Simple moving average with momentum
ma_short = df['Close'].rolling(window=10).mean().iloc[-1]
ma_long = df['Close'].rolling(window=30).mean().iloc[-1]
momentum = (ma_short - ma_long) / ma_long
last_price = df['Close'].iloc[-1]
stat_pred = []
for i in range(1, prediction_days + 1):
# Mean reversion with momentum
prediction = last_price * (1 + momentum * 0.5 - 0.001 * i) # Decay factor
stat_pred.append(prediction)
predictions["statistical"] = np.array(stat_pred)
uncertainties["statistical"] = np.array([returns.std() * last_price * np.sqrt(i) for i in range(1, prediction_days + 1)])
else:
predictions["statistical"] = np.array([])
uncertainties["statistical"] = np.array([])
except Exception as e:
print(f"Statistical prediction error: {str(e)}")
predictions["statistical"] = np.array([])
uncertainties["statistical"] = np.array([])
# Combine predictions
valid_predictions = {k: v for k, v in predictions.items() if len(v) > 0}
valid_uncertainties = {k: v for k, v in uncertainties.items() if len(v) > 0}
if not valid_predictions:
return np.array([]), np.array([])
# Weighted ensemble
total_weight = sum(ensemble_weights.get(k, 0) for k in valid_predictions.keys())
if total_weight == 0:
return np.array([]), np.array([])
# Normalize weights
normalized_weights = {k: ensemble_weights.get(k, 0) / total_weight for k in valid_predictions.keys()}
# Calculate weighted mean and uncertainty
max_length = max(len(v) for v in valid_predictions.values())
ensemble_mean = np.zeros(max_length)
ensemble_uncertainty = np.zeros(max_length)
for model, pred in valid_predictions.items():
weight = normalized_weights[model]
if len(pred) < max_length:
# Extend prediction using last value
extended_pred = np.concatenate([pred, np.full(max_length - len(pred), pred[-1])])
extended_unc = np.concatenate([valid_uncertainties[model], np.full(max_length - len(pred), valid_uncertainties[model][-1])])
else:
extended_pred = pred[:max_length]
extended_unc = valid_uncertainties[model][:max_length]
ensemble_mean += weight * extended_pred
ensemble_uncertainty += weight * extended_unc
return ensemble_mean, ensemble_uncertainty
def stress_test_scenarios(df: pd.DataFrame, prediction: np.ndarray,
scenarios: Dict = None) -> Dict:
"""
Perform stress testing under various market scenarios.
Args:
df (pd.DataFrame): Historical data
prediction (np.ndarray): Base prediction
scenarios (Dict): Stress test scenarios
Returns:
Dict: Stress test results
"""
if scenarios is None:
scenarios = {
"market_crash": {"volatility_multiplier": 3.0, "return_shock": -0.15},
"high_volatility": {"volatility_multiplier": 2.0, "return_shock": -0.05},
"low_volatility": {"volatility_multiplier": 0.5, "return_shock": 0.02},
"bull_market": {"volatility_multiplier": 1.2, "return_shock": 0.10},
"interest_rate_shock": {"volatility_multiplier": 1.5, "return_shock": -0.08}
}
base_volatility = df['Volatility'].iloc[-1]
base_return = df['Returns'].mean()
last_price = df['Close'].iloc[-1]
stress_results = {}
for scenario_name, params in scenarios.items():
try:
# Calculate stressed parameters
stressed_vol = base_volatility * params["volatility_multiplier"]
stressed_return = base_return + params["return_shock"]
# Generate stressed prediction
stressed_pred = []
for i, pred in enumerate(prediction):
# Apply stress factors
stress_factor = 1 + stressed_return * (i + 1) / 252
volatility_impact = np.random.normal(0, stressed_vol * np.sqrt((i + 1) / 252))
stressed_price = pred * stress_factor * (1 + volatility_impact)
stressed_pred.append(stressed_price)
# Calculate stress metrics
stress_results[scenario_name] = {
"prediction": np.array(stressed_pred),
"max_loss": min(stressed_pred) / last_price - 1,
"volatility": stressed_vol,
"expected_return": stressed_return,
"var_95": np.percentile([p / last_price - 1 for p in stressed_pred], 5)
}
except Exception as e:
print(f"Stress test error for {scenario_name}: {str(e)}")
stress_results[scenario_name] = {"error": str(e)}
return stress_results
def calculate_skewed_uncertainty(quantiles: np.ndarray, confidence_level: float = 0.9) -> np.ndarray:
"""
Calculate uncertainty accounting for skewness in return distributions.
Args:
quantiles (np.ndarray): Quantile predictions from Chronos
confidence_level (float): Confidence level for uncertainty calculation
Returns:
np.ndarray: Uncertainty estimates
"""
try:
lower = quantiles[0, :, 0]
median = quantiles[0, :, 1]
upper = quantiles[0, :, 2]
# Calculate skewness for each prediction point
uncertainties = []
for i in range(len(lower)):
# Calculate skewness
if upper[i] != median[i] and median[i] != lower[i]:
skewness = (median[i] - lower[i]) / (upper[i] - median[i])
else:
skewness = 1.0
# Adjust z-score based on skewness
if skewness > 1.2: # Right-skewed
z_score = stats.norm.ppf(confidence_level) * (1 + 0.1 * skewness)
elif skewness < 0.8: # Left-skewed
z_score = stats.norm.ppf(confidence_level) * (1 - 0.1 * abs(skewness))
else:
z_score = stats.norm.ppf(confidence_level)
# Calculate uncertainty
uncertainty = (upper[i] - lower[i]) / (2 * z_score)
uncertainties.append(uncertainty)
return np.array(uncertainties)
except Exception as e:
print(f"Skewed uncertainty calculation error: {str(e)}")
# Fallback to simple calculation
return (quantiles[0, :, 2] - quantiles[0, :, 0]) / (2 * 1.645)
def adaptive_smoothing(new_pred: np.ndarray, historical_pred: np.ndarray,
prediction_uncertainty: np.ndarray) -> np.ndarray:
"""
Apply adaptive smoothing based on prediction uncertainty.
Args:
new_pred (np.ndarray): New predictions
historical_pred (np.ndarray): Historical predictions
prediction_uncertainty (np.ndarray): Prediction uncertainty
Returns:
np.ndarray: Smoothed predictions
"""
try:
if len(historical_pred) == 0:
return new_pred
# Calculate adaptive alpha based on uncertainty
uncertainty_ratio = prediction_uncertainty / np.mean(np.abs(historical_pred))
if uncertainty_ratio > 0.1: # High uncertainty
alpha = 0.1 # More smoothing
elif uncertainty_ratio < 0.05: # Low uncertainty
alpha = 0.5 # Less smoothing
else:
alpha = 0.3 # Default
# Apply weighted smoothing
smoothed = alpha * new_pred + (1 - alpha) * historical_pred[-len(new_pred):]
return smoothed
except Exception as e:
print(f"Adaptive smoothing error: {str(e)}")
return new_pred
def advanced_trading_signals(df: pd.DataFrame, regime_info: Dict = None) -> Dict:
"""
Generate advanced trading signals with confidence levels and regime awareness.
Args:
df (pd.DataFrame): Stock data
regime_info (Dict): Market regime information
Returns:
Dict: Advanced trading signals
"""
try:
# Calculate signal strength and confidence
rsi = df['RSI'].iloc[-1]
macd = df['MACD'].iloc[-1]
macd_signal = df['MACD_Signal'].iloc[-1]
rsi_strength = abs(rsi - 50) / 50 # 0-1 scale
macd_strength = abs(macd - macd_signal) / df['Close'].iloc[-1]
# Regime-adjusted thresholds
if regime_info and "volatilities" in regime_info:
volatility_regime = df['Volatility'].iloc[-1] / np.mean(regime_info["volatilities"])
else:
volatility_regime = 1.0
# Adjust RSI thresholds based on volatility
rsi_oversold = 30 + (volatility_regime - 1) * 10
rsi_overbought = 70 - (volatility_regime - 1) * 10
# Calculate signals with confidence
signals = {}
# RSI signal
if rsi < rsi_oversold:
rsi_signal = "Oversold"
rsi_confidence = min(0.9, 0.5 + rsi_strength * 0.4)
elif rsi > rsi_overbought:
rsi_signal = "Overbought"
rsi_confidence = min(0.9, 0.5 + rsi_strength * 0.4)
else:
rsi_signal = "Neutral"
rsi_confidence = 0.3
signals["RSI"] = {
"signal": rsi_signal,
"strength": rsi_strength,
"confidence": rsi_confidence,
"value": rsi
}
# MACD signal
if macd > macd_signal:
macd_signal = "Buy"
macd_confidence = min(0.8, 0.4 + macd_strength * 40)
else:
macd_signal = "Sell"
macd_confidence = min(0.8, 0.4 + macd_strength * 40)
signals["MACD"] = {
"signal": macd_signal,
"strength": macd_strength,
"confidence": macd_confidence,
"value": macd
}
# Bollinger Bands signal
if 'BB_Upper' in df.columns and 'BB_Lower' in df.columns:
current_price = df['Close'].iloc[-1]
bb_upper = df['BB_Upper'].iloc[-1]
bb_lower = df['BB_Lower'].iloc[-1]
# Calculate position within Bollinger Bands (0-1 scale)
bb_position = (current_price - bb_lower) / (bb_upper - bb_lower) if bb_upper != bb_lower else 0.5
bb_strength = abs(bb_position - 0.5) * 2 # 0-1 scale, strongest at edges
if current_price < bb_lower:
bb_signal = "Buy"
bb_confidence = 0.7
elif current_price > bb_upper:
bb_signal = "Sell"
bb_confidence = 0.7
else:
bb_signal = "Hold"
bb_confidence = 0.5
signals["Bollinger"] = {
"signal": bb_signal,
"strength": bb_strength,
"confidence": bb_confidence,
"position": bb_position
}
# SMA signal
if 'SMA_20' in df.columns and 'SMA_50' in df.columns:
sma_20 = df['SMA_20'].iloc[-1]
sma_50 = df['SMA_50'].iloc[-1]
# Calculate SMA strength based on ratio
sma_ratio = sma_20 / sma_50 if sma_50 != 0 else 1.0
sma_strength = abs(sma_ratio - 1.0) # 0-1 scale, strongest when ratio differs most from 1
if sma_20 > sma_50:
sma_signal = "Buy"
sma_confidence = 0.6
else:
sma_signal = "Sell"
sma_confidence = 0.6
signals["SMA"] = {
"signal": sma_signal,
"strength": sma_strength,
"confidence": sma_confidence,
"ratio": sma_ratio
}
# Calculate weighted overall signal
buy_signals = []
sell_signals = []
for signal_name, signal_data in signals.items():
# Get strength with default value if not present
strength = signal_data.get("strength", 0.5) # Default strength of 0.5
confidence = signal_data.get("confidence", 0.5) # Default confidence of 0.5
if signal_data["signal"] == "Buy":
buy_signals.append(strength * confidence)
elif signal_data["signal"] == "Sell":
sell_signals.append(strength * confidence)
weighted_buy = sum(buy_signals) if buy_signals else 0
weighted_sell = sum(sell_signals) if sell_signals else 0
if weighted_buy > weighted_sell:
overall_signal = "Buy"
overall_confidence = weighted_buy / (weighted_buy + weighted_sell) if (weighted_buy + weighted_sell) > 0 else 0
elif weighted_sell > weighted_buy:
overall_signal = "Sell"
overall_confidence = weighted_sell / (weighted_buy + weighted_sell) if (weighted_buy + weighted_sell) > 0 else 0
else:
overall_signal = "Hold"
overall_confidence = 0.5
return {
"signals": signals,
"overall_signal": overall_signal,
"confidence": overall_confidence,
"regime_adjusted": regime_info is not None
}
except Exception as e:
print(f"Advanced trading signals error: {str(e)}")
return {"error": str(e)}
def apply_financial_smoothing(data: np.ndarray, smoothing_type: str = "exponential",
window_size: int = 5, alpha: float = 0.3,
poly_order: int = 3, use_smoothing: bool = True) -> np.ndarray:
"""
Apply financial smoothing algorithms to time series data.
Args:
data (np.ndarray): Input time series data
smoothing_type (str): Type of smoothing to apply
- 'exponential': Exponential moving average (good for trend following)
- 'moving_average': Simple moving average (good for noise reduction)
- 'kalman': Kalman filter (good for adaptive smoothing)
- 'savitzky_golay': Savitzky-Golay filter (good for preserving peaks/valleys)
- 'double_exponential': Double exponential smoothing (good for trend + seasonality)
- 'triple_exponential': Triple exponential smoothing (Holt-Winters, good for complex patterns)
- 'adaptive': Adaptive smoothing based on volatility
- 'none': No smoothing applied
window_size (int): Window size for moving average and Savitzky-Golay
alpha (float): Smoothing factor for exponential methods (0-1)
poly_order (int): Polynomial order for Savitzky-Golay filter
use_smoothing (bool): Whether to apply smoothing
Returns:
np.ndarray: Smoothed data
"""
if not use_smoothing or smoothing_type == "none" or len(data) < 3:
return data
try:
if smoothing_type == "exponential":
# Exponential Moving Average - good for trend following
smoothed = np.zeros_like(data)
smoothed[0] = data[0]
for i in range(1, len(data)):
smoothed[i] = alpha * data[i] + (1 - alpha) * smoothed[i-1]
return smoothed
elif smoothing_type == "moving_average":
# Simple Moving Average - good for noise reduction
if len(data) < window_size:
return data
smoothed = np.zeros_like(data)
# Handle the beginning of the series
for i in range(min(window_size - 1, len(data))):
smoothed[i] = np.mean(data[:i+1])
# Apply moving average for the rest
for i in range(window_size - 1, len(data)):
smoothed[i] = np.mean(data[i-window_size+1:i+1])
return smoothed
elif smoothing_type == "kalman":
# Kalman Filter - adaptive smoothing
if len(data) < 2:
return data
# Initialize Kalman filter parameters
Q = 0.01 # Process noise
R = 0.1 # Measurement noise
P = 1.0 # Initial estimate error
x = data[0] # Initial state estimate
smoothed = np.zeros_like(data)
smoothed[0] = x
for i in range(1, len(data)):
# Prediction step
x_pred = x
P_pred = P + Q
# Update step
K = P_pred / (P_pred + R) # Kalman gain
x = x_pred + K * (data[i] - x_pred)
P = (1 - K) * P_pred
smoothed[i] = x
return smoothed
elif smoothing_type == "savitzky_golay":
# Savitzky-Golay filter - preserves peaks and valleys
if len(data) < window_size:
return data
# Ensure window_size is odd
if window_size % 2 == 0:
window_size += 1
# Ensure polynomial order is less than window_size
if poly_order >= window_size:
poly_order = window_size - 1
try:
from scipy.signal import savgol_filter
return savgol_filter(data, window_size, poly_order)
except ImportError:
# Fallback to simple moving average if scipy not available
return apply_financial_smoothing(data, "moving_average", window_size)
elif smoothing_type == "double_exponential":
# Double Exponential Smoothing (Holt's method) - trend + level
if len(data) < 3:
return data
smoothed = np.zeros_like(data)
trend = np.zeros_like(data)
# Initialize
smoothed[0] = data[0]
trend[0] = data[1] - data[0] if len(data) > 1 else 0
# Apply double exponential smoothing
for i in range(1, len(data)):
prev_smoothed = smoothed[i-1]
prev_trend = trend[i-1]
smoothed[i] = alpha * data[i] + (1 - alpha) * (prev_smoothed + prev_trend)
trend[i] = alpha * (smoothed[i] - prev_smoothed) + (1 - alpha) * prev_trend
return smoothed
elif smoothing_type == "triple_exponential":
# Triple Exponential Smoothing (Holt-Winters) - trend + level + seasonality
if len(data) < 6:
return apply_financial_smoothing(data, "double_exponential", window_size, alpha)
# For simplicity, we'll use a seasonal period of 5 (common for financial data)
season_period = min(5, len(data) // 2)
smoothed = np.zeros_like(data)
trend = np.zeros_like(data)
season = np.zeros_like(data)
# Initialize
smoothed[0] = data[0]
trend[0] = (data[season_period] - data[0]) / season_period if len(data) > season_period else 0
# Initialize seasonal components
for i in range(season_period):
season[i] = data[i] - smoothed[0]
# Apply triple exponential smoothing
for i in range(1, len(data)):
prev_smoothed = smoothed[i-1]
prev_trend = trend[i-1]
prev_season = season[(i-1) % season_period]
smoothed[i] = alpha * (data[i] - prev_season) + (1 - alpha) * (prev_smoothed + prev_trend)
trend[i] = alpha * (smoothed[i] - prev_smoothed) + (1 - alpha) * prev_trend
season[i % season_period] = alpha * (data[i] - smoothed[i]) + (1 - alpha) * prev_season
return smoothed
elif smoothing_type == "adaptive":
# Adaptive smoothing based on volatility
if len(data) < 5:
return data
# Calculate rolling volatility
returns = np.diff(data) / data[:-1]
volatility = np.zeros_like(data)
volatility[0] = np.std(returns) if len(returns) > 0 else 0.01
for i in range(1, len(data)):
if i < 5:
volatility[i] = np.std(returns[:i]) if i > 0 else 0.01
else:
volatility[i] = np.std(returns[i-5:i])
# Normalize volatility to smoothing factor
vol_factor = np.clip(volatility / np.mean(volatility), 0.1, 0.9)
adaptive_alpha = 1 - vol_factor # Higher volatility = less smoothing
# Apply adaptive exponential smoothing
smoothed = np.zeros_like(data)
smoothed[0] = data[0]
for i in range(1, len(data)):
current_alpha = adaptive_alpha[i]
smoothed[i] = current_alpha * data[i] + (1 - current_alpha) * smoothed[i-1]
return smoothed
else:
# Default to exponential smoothing
return apply_financial_smoothing(data, "exponential", window_size, alpha)
except Exception as e:
print(f"Smoothing error: {str(e)}")
return data
def create_interface():
"""Create the Gradio interface with separate tabs for different timeframes"""
with gr.Blocks(title="Advanced Stock Prediction Analysis") as demo:
gr.Markdown("# Advanced Stock Prediction Analysis")
gr.Markdown("Analyze stocks with advanced features including regime detection, ensemble methods, and stress testing.")
# Add market status message
market_status = "Market is currently closed" if not is_market_open() else "Market is currently open"
next_trading_day = get_next_trading_day()
gr.Markdown(f"""
### Market Status: {market_status}
Next trading day: {next_trading_day.strftime('%Y-%m-%d')}
""")
# Advanced Settings Accordion
with gr.Accordion("Advanced Settings", open=False):
with gr.Row():
with gr.Column():
use_ensemble = gr.Checkbox(label="Use Ensemble Methods", value=True)
use_regime_detection = gr.Checkbox(label="Use Regime Detection", value=True)
use_stress_testing = gr.Checkbox(label="Use Stress Testing", value=True)
use_smoothing = gr.Checkbox(label="Use Smoothing", value=True)
smoothing_type = gr.Dropdown(
choices=["exponential", "moving_average", "kalman", "savitzky_golay",
"double_exponential", "triple_exponential", "adaptive", "none"],
label="Smoothing Type",
value="exponential",
info="""Smoothing algorithms:
• Exponential: Trend following (default)
• Moving Average: Noise reduction
• Kalman: Adaptive smoothing
• Savitzky-Golay: Preserves peaks/valleys
• Double Exponential: Trend + level
• Triple Exponential: Complex patterns
• Adaptive: Volatility-based
• None: No smoothing"""
)
smoothing_window = gr.Slider(
minimum=3,
maximum=21,
value=5,
step=1,
label="Smoothing Window Size",
info="Window size for moving average and Savitzky-Golay filters"
)
smoothing_alpha = gr.Slider(
minimum=0.1,
maximum=0.9,
value=0.3,
step=0.05,
label="Smoothing Alpha",
info="Smoothing factor for exponential methods (0.1-0.9)"
)
risk_free_rate = gr.Slider(
minimum=0.0,
maximum=0.1,
value=0.02,
step=0.001,
label="Risk-Free Rate (Annual)"
)
market_index = gr.Dropdown(
choices=["^GSPC", "^DJI", "^IXIC", "^RUT"],
label="Market Index for Correlation",
value="^GSPC"
)
random_real_points = gr.Slider(
minimum=0,
maximum=16,
value=4,
step=1,
label="Random Real Points in Long-Horizon Context"
)
with gr.Column():
gr.Markdown("### Ensemble Weights")
chronos_weight = gr.Slider(
minimum=0.0,
maximum=1.0,
value=0.6,
step=0.1,
label="Chronos Weight"
)
technical_weight = gr.Slider(
minimum=0.0,
maximum=1.0,
value=0.2,
step=0.1,
label="Technical Weight"
)
statistical_weight = gr.Slider(
minimum=0.0,
maximum=1.0,
value=0.2,
step=0.1,
label="Statistical Weight"
)
with gr.Tabs() as tabs:
# Daily Analysis Tab
with gr.TabItem("Daily Analysis"):
with gr.Row():
with gr.Column():
daily_symbol = gr.Textbox(label="Stock Symbol (e.g., AAPL)", value="AAPL")
daily_prediction_days = gr.Slider(
minimum=1,
maximum=365,
value=30,
step=1,
label="Days to Predict"
)
daily_lookback_days = gr.Slider(
minimum=1,
maximum=3650,
value=365,
step=1,
label="Historical Lookback (Days)"
)
daily_strategy = gr.Dropdown(
choices=["chronos", "technical"],
label="Prediction Strategy",
value="chronos"
)
daily_predict_btn = gr.Button("Analyze Stock")
gr.Markdown("""
**Daily Analysis Features:**
- **Extended Data Range**: Up to 10 years of historical data (3650 days)
- **24/7 Availability**: Available regardless of market hours
- **Auto-Adjusted Data**: Automatically adjusted for splits and dividends
- **Comprehensive Financial Ratios**: P/E, PEG, Price-to-Book, Price-to-Sales, and more
- **Advanced Risk Metrics**: Sharpe ratio, VaR, drawdown analysis, market correlation
- **Market Regime Detection**: Identifies bull/bear/sideways market conditions
- **Stress Testing**: Scenario analysis under various market conditions
- **Ensemble Methods**: Combines multiple prediction models for improved accuracy
- **Maximum prediction period**: 365 days
- **Ideal for**: Medium to long-term investment analysis, portfolio management, and strategic planning
- **Technical Indicators**: RSI, MACD, Bollinger Bands, moving averages optimized for daily data
- **Volume Analysis**: Average daily volume, volume volatility, and liquidity metrics
- **Sector Analysis**: Industry classification, market cap ranking, and sector-specific metrics
""")
with gr.Column():
daily_plot = gr.Plot(label="Analysis and Prediction")
with gr.Row():
with gr.Column():
gr.Markdown("### Structured Product Metrics")
daily_metrics = gr.JSON(label="Product Metrics")
gr.Markdown("### Advanced Risk Analysis")
daily_risk_metrics = gr.JSON(label="Risk Metrics")
gr.Markdown("### Market Regime Analysis")
daily_regime_metrics = gr.JSON(label="Regime Metrics")
gr.Markdown("### Trading Signals")
daily_signals = gr.JSON(label="Trading Signals")
gr.Markdown("### Advanced Trading Signals")
daily_signals_advanced = gr.JSON(label="Advanced Trading Signals")
with gr.Column():
gr.Markdown("### Sector & Financial Analysis")
daily_sector_metrics = gr.JSON(label="Sector Metrics")
gr.Markdown("### Stress Test Results")
daily_stress_results = gr.JSON(label="Stress Test Results")
gr.Markdown("### Ensemble Analysis")
daily_ensemble_metrics = gr.JSON(label="Ensemble Metrics")
# Hourly Analysis Tab
with gr.TabItem("Hourly Analysis"):
with gr.Row():
with gr.Column():
hourly_symbol = gr.Textbox(label="Stock Symbol (e.g., AAPL)", value="AAPL")
hourly_prediction_days = gr.Slider(
minimum=1,
maximum=7, # Limited to 7 days for hourly predictions
value=3,
step=1,
label="Days to Predict"
)
hourly_lookback_days = gr.Slider(
minimum=1,
maximum=60, # Enhanced to 60 days for hourly data
value=14,
step=1,
label="Historical Lookback (Days)"
)
hourly_strategy = gr.Dropdown(
choices=["chronos", "technical"],
label="Prediction Strategy",
value="chronos"
)
hourly_predict_btn = gr.Button("Analyze Stock")
gr.Markdown("""
**Hourly Analysis Features:**
- **Extended Data Range**: Up to 60 days of historical data
- **Pre/Post Market Data**: Includes extended hours trading data
- **Auto-Adjusted Data**: Automatically adjusted for splits and dividends
- **Metrics**: Intraday volatility, volume analysis, and momentum indicators
- **Comprehensive Financial Ratios**: P/E, PEG, Price-to-Book, and more
- **Maximum prediction period**: 7 days
- **Data available during market hours only**
""")
with gr.Column():
hourly_plot = gr.Plot(label="Analysis and Prediction")
hourly_signals = gr.JSON(label="Trading Signals")
with gr.Row():
with gr.Column():
gr.Markdown("### Structured Product Metrics")
hourly_metrics = gr.JSON(label="Product Metrics")
gr.Markdown("### Advanced Risk Analysis")
hourly_risk_metrics = gr.JSON(label="Risk Metrics")
gr.Markdown("### Market Regime Analysis")
hourly_regime_metrics = gr.JSON(label="Regime Metrics")
gr.Markdown("### Trading Signals")
hourly_signals_advanced = gr.JSON(label="Advanced Trading Signals")
with gr.Column():
gr.Markdown("### Sector & Financial Analysis")
hourly_sector_metrics = gr.JSON(label="Sector Metrics")
gr.Markdown("### Stress Test Results")
hourly_stress_results = gr.JSON(label="Stress Test Results")
gr.Markdown("### Ensemble Analysis")
hourly_ensemble_metrics = gr.JSON(label="Ensemble Metrics")
# 15-Minute Analysis Tab
with gr.TabItem("15-Minute Analysis"):
with gr.Row():
with gr.Column():
min15_symbol = gr.Textbox(label="Stock Symbol (e.g., AAPL)", value="AAPL")
min15_prediction_days = gr.Slider(
minimum=1,
maximum=2, # Limited to 2 days for 15-minute predictions
value=1,
step=1,
label="Days to Predict"
)
min15_lookback_days = gr.Slider(
minimum=1,
maximum=7, # 7 days for 15-minute data
value=3,
step=1,
label="Historical Lookback (Days)"
)
min15_strategy = gr.Dropdown(
choices=["chronos", "technical"],
label="Prediction Strategy",
value="chronos"
)
min15_predict_btn = gr.Button("Analyze Stock")
gr.Markdown("""
**15-Minute Analysis Features:**
- **Data Range**: Up to 7 days of historical data (vs 5 days previously)
- **High-Frequency Metrics**: Intraday volatility, volume-price trends, momentum analysis
- **Pre/Post Market Data**: Includes extended hours trading data
- **Auto-Adjusted Data**: Automatically adjusted for splits and dividends
- **Enhanced Technical Indicators**: Optimized for short-term trading
- **Maximum prediction period**: 2 days
- **Requires at least 64 data points for Chronos predictions**
- **Data available during market hours only**
""")
with gr.Column():
min15_plot = gr.Plot(label="Analysis and Prediction")
min15_signals = gr.JSON(label="Trading Signals")
with gr.Row():
with gr.Column():
gr.Markdown("### Structured Product Metrics")
min15_metrics = gr.JSON(label="Product Metrics")
gr.Markdown("### Advanced Risk Analysis")
min15_risk_metrics = gr.JSON(label="Risk Metrics")
gr.Markdown("### Market Regime Analysis")
min15_regime_metrics = gr.JSON(label="Regime Metrics")
gr.Markdown("### Trading Signals")
min15_signals_advanced = gr.JSON(label="Advanced Trading Signals")
with gr.Column():
gr.Markdown("### Sector & Financial Analysis")
min15_sector_metrics = gr.JSON(label="Sector Metrics")
gr.Markdown("### Stress Test Results")
min15_stress_results = gr.JSON(label="Stress Test Results")
gr.Markdown("### Ensemble Analysis")
min15_ensemble_metrics = gr.JSON(label="Ensemble Metrics")
def analyze_stock(symbol, timeframe, prediction_days, lookback_days, strategy,
use_ensemble, use_regime_detection, use_stress_testing,
risk_free_rate, market_index, chronos_weight, technical_weight, statistical_weight,
random_real_points, use_smoothing, smoothing_type, smoothing_window, smoothing_alpha):
try:
# Create ensemble weights
ensemble_weights = {
"chronos": chronos_weight,
"technical": technical_weight,
"statistical": statistical_weight
}
# Get market data for correlation analysis
market_df = get_market_data(market_index, lookback_days)
market_returns = market_df['Returns'] if not market_df.empty else None
# Make prediction with advanced features
signals, fig = make_prediction(
symbol=symbol,
timeframe=timeframe,
prediction_days=prediction_days,
strategy=strategy,
use_ensemble=use_ensemble,
use_regime_detection=use_regime_detection,
use_stress_testing=use_stress_testing,
risk_free_rate=risk_free_rate,
ensemble_weights=ensemble_weights,
market_index=market_index,
random_real_points=random_real_points,
use_smoothing=use_smoothing,
smoothing_type=smoothing_type,
smoothing_window=smoothing_window,
smoothing_alpha=smoothing_alpha
)
# Get historical data for additional metrics
df = get_historical_data(symbol, timeframe, lookback_days)
# Calculate structured product metrics
product_metrics = {
"Market_Cap": df['Market_Cap'].iloc[-1],
"Sector": df['Sector'].iloc[-1],
"Industry": df['Industry'].iloc[-1],
"Dividend_Yield": df['Dividend_Yield'].iloc[-1],
"Avg_Daily_Volume": df['Avg_Daily_Volume'].iloc[-1],
"Volume_Volatility": df['Volume_Volatility'].iloc[-1],
"Enterprise_Value": df['Enterprise_Value'].iloc[-1],
"P/E_Ratio": df['P/E_Ratio'].iloc[-1],
"Forward_P/E": df['Forward_P/E'].iloc[-1],
"PEG_Ratio": df['PEG_Ratio'].iloc[-1],
"Price_to_Book": df['Price_to_Book'].iloc[-1],
"Price_to_Sales": df['Price_to_Sales'].iloc[-1]
}
# Calculate advanced risk metrics
risk_metrics = calculate_advanced_risk_metrics(df, market_returns, risk_free_rate)
# Calculate sector metrics
sector_metrics = {
"Sector": df['Sector'].iloc[-1],
"Industry": df['Industry'].iloc[-1],
"Market_Cap_Rank": "Large" if df['Market_Cap'].iloc[-1] > 1e10 else "Mid" if df['Market_Cap'].iloc[-1] > 1e9 else "Small",
"Liquidity_Score": "High" if df['Avg_Daily_Volume'].iloc[-1] > 1e6 else "Medium" if df['Avg_Daily_Volume'].iloc[-1] > 1e5 else "Low",
"Gross_Margin": df['Gross_Margin'].iloc[-1],
"Operating_Margin": df['Operating_Margin'].iloc[-1],
"Net_Margin": df['Net_Margin'].iloc[-1]
}
# Add intraday-specific metrics for shorter timeframes
if timeframe in ["1h", "15m"]:
intraday_metrics = {
"Intraday_Volatility": df['Intraday_Volatility'].iloc[-1] if 'Intraday_Volatility' in df.columns else 0,
"Volume_Ratio": df['Volume_Ratio'].iloc[-1] if 'Volume_Ratio' in df.columns else 0,
"Price_Momentum": df['Price_Momentum'].iloc[-1] if 'Price_Momentum' in df.columns else 0,
"Volume_Momentum": df['Volume_Momentum'].iloc[-1] if 'Volume_Momentum' in df.columns else 0,
"Volume_Price_Trend": df['Volume_Price_Trend'].iloc[-1] if 'Volume_Price_Trend' in df.columns else 0
}
product_metrics.update(intraday_metrics)
# Extract regime and stress test information
regime_metrics = signals.get("regime_info", {})
stress_results = signals.get("stress_test_results", {})
ensemble_metrics = {
"ensemble_used": signals.get("ensemble_used", False),
"ensemble_weights": ensemble_weights
}
# Separate basic and advanced signals
basic_signals = {
"RSI": signals.get("RSI", "Neutral"),
"MACD": signals.get("MACD", "Hold"),
"Bollinger": signals.get("Bollinger", "Hold"),
"SMA": signals.get("SMA", "Hold"),
"Overall": signals.get("Overall", "Hold"),
"symbol": signals.get("symbol", symbol),
"timeframe": signals.get("timeframe", timeframe),
"strategy_used": signals.get("strategy_used", strategy)
}
advanced_signals = signals.get("advanced_signals", {})
return basic_signals, fig, product_metrics, risk_metrics, sector_metrics, regime_metrics, stress_results, ensemble_metrics, advanced_signals
except Exception as e:
error_message = str(e)
if "Market is currently closed" in error_message:
error_message = f"{error_message}. Please try again during market hours or use daily timeframe."
elif "Insufficient data points" in error_message:
error_message = f"Not enough data available for {symbol} in {timeframe} timeframe. Please try a different timeframe or symbol."
elif "no price data found" in error_message:
error_message = f"No data available for {symbol} in {timeframe} timeframe. Please try a different timeframe or symbol."
raise gr.Error(error_message)
# Daily analysis button click
def daily_analysis(s: str, pd: int, ld: int, st: str, ue: bool, urd: bool, ust: bool,
rfr: float, mi: str, cw: float, tw: float, sw: float,
rrp: int, usm: bool, smt: str, sww: float, sa: float) -> Tuple[Dict, go.Figure, Dict, Dict, Dict, Dict, Dict, Dict, Dict]:
"""
Process daily timeframe stock analysis with advanced features.
This function performs comprehensive stock analysis using daily data with support for
multiple prediction strategies, ensemble methods, regime detection, and stress testing.
It's designed for medium to long-term investment analysis with up to 365 days of prediction.
Args:
s (str): Stock symbol (e.g., "AAPL", "MSFT", "GOOGL", "TSLA")
Must be a valid stock symbol available on Yahoo Finance
pd (int): Number of days to predict (1-365)
The forecast horizon for the analysis. Longer periods may have higher uncertainty
ld (int): Historical lookback period in days (1-3650)
Amount of historical data to use for analysis. More data generally improves accuracy
st (str): Prediction strategy to use ("chronos" or "technical")
- "chronos": Uses Amazon's Chronos T5 model for time series forecasting
- "technical": Uses traditional technical analysis indicators
ue (bool): Use ensemble methods
When True, combines multiple prediction models for improved accuracy
urd (bool): Use regime detection
When True, detects market regimes (bull/bear/sideways) to adjust predictions
ust (bool): Use stress testing
When True, performs scenario analysis under various market conditions
rfr (float): Risk-free rate (0.0-0.1)
Annual risk-free rate used for risk-adjusted return calculations
mi (str): Market index for correlation analysis
Options: "^GSPC" (S&P 500), "^DJI" (Dow Jones), "^IXIC" (NASDAQ), "^RUT" (Russell 2000)
cw (float): Chronos weight in ensemble (0.0-1.0)
Weight given to Chronos model predictions in ensemble methods
tw (float): Technical weight in ensemble (0.0-1.0)
Weight given to technical analysis predictions in ensemble methods
sw (float): Statistical weight in ensemble (0.0-1.0)
Weight given to statistical model predictions in ensemble methods
rrp (int): Number of random real points to include in long-horizon context
usm (bool): Use smoothing
When True, applies smoothing to predictions to reduce noise and improve continuity
smt (str): Smoothing type to use
Options: "exponential", "moving_average", "kalman", "savitzky_golay", "double_exponential", "triple_exponential", "adaptive", "none"
sww (float): Smoothing window size for moving average and Savitzky-Golay
sa (float): Smoothing alpha for exponential methods (0.1-0.9)
Returns:
Tuple[Dict, go.Figure, Dict, Dict, Dict, Dict, Dict, Dict, Dict]: Analysis results containing:
- Dict: Basic trading signals (RSI, MACD, Bollinger Bands, SMA, Overall)
- go.Figure: Interactive plot with historical data, predictions, and confidence intervals
- Dict: Structured product metrics (Market Cap, P/E ratios, financial ratios)
- Dict: Advanced risk metrics (Sharpe ratio, VaR, drawdown, correlation)
- Dict: Sector and industry analysis metrics
- Dict: Market regime detection results
- Dict: Stress testing scenario results
- Dict: Ensemble method configuration and results
- Dict: Advanced trading signals with confidence levels
Raises:
gr.Error: If data cannot be fetched, insufficient data points, or other analysis errors
Common errors include invalid symbols, market closure, or insufficient historical data
Example:
>>> signals, plot, metrics, risk, sector, regime, stress, ensemble, advanced = daily_analysis(
... "AAPL", 30, 365, "chronos", True, True, True, 0.02, "^GSPC", 0.6, 0.2, 0.2, 4, True, "exponential", 5, 0.3
... )
Notes:
- Daily analysis is available 24/7 regardless of market hours
- Maximum prediction period is 365 days
- Historical data can go back up to 10 years (3650 days)
- Ensemble weights should sum to 1.0 for optimal results
- Risk-free rate is typically between 0.02-0.05 (2-5% annually)
- Smoothing helps reduce prediction noise but may reduce responsiveness to sudden changes
"""
return analyze_stock(s, "1d", pd, ld, st, ue, urd, ust, rfr, mi, cw, tw, sw, rrp, usm, smt, sww, sa)
daily_predict_btn.click(
fn=daily_analysis,
inputs=[daily_symbol, daily_prediction_days, daily_lookback_days, daily_strategy,
use_ensemble, use_regime_detection, use_stress_testing, risk_free_rate, market_index,
chronos_weight, technical_weight, statistical_weight,
random_real_points, use_smoothing, smoothing_type, smoothing_window, smoothing_alpha],
outputs=[daily_signals, daily_plot, daily_metrics, daily_risk_metrics, daily_sector_metrics,
daily_regime_metrics, daily_stress_results, daily_ensemble_metrics, daily_signals_advanced]
)
# Hourly analysis button click
def hourly_analysis(s: str, pd: int, ld: int, st: str, ue: bool, urd: bool, ust: bool,
rfr: float, mi: str, cw: float, tw: float, sw: float,
rrp: int, usm: bool, smt: str, sww: float, sa: float) -> Tuple[Dict, go.Figure, Dict, Dict, Dict, Dict, Dict, Dict, Dict]:
"""
Process hourly timeframe stock analysis with advanced features.
This function performs high-frequency stock analysis using hourly data, ideal for
short to medium-term trading strategies. It includes intraday volatility analysis,
volume-price trends, and momentum indicators optimized for hourly timeframes.
Args:
s (str): Stock symbol (e.g., "AAPL", "MSFT", "GOOGL", "TSLA")
Must be a valid stock symbol with sufficient liquidity for hourly analysis
pd (int): Number of days to predict (1-7)
Limited to 7 days due to Yahoo Finance hourly data constraints
ld (int): Historical lookback period in days (1-60)
Enhanced to 60 days for hourly data (vs standard 30 days)
st (str): Prediction strategy to use ("chronos" or "technical")
- "chronos": Uses Amazon's Chronos T5 model optimized for hourly data
- "technical": Uses technical indicators adjusted for hourly timeframes
ue (bool): Use ensemble methods
Combines multiple models for improved short-term prediction accuracy
urd (bool): Use regime detection
Detects intraday market regimes and volatility patterns
ust (bool): Use stress testing
Performs scenario analysis for short-term market shocks
rfr (float): Risk-free rate (0.0-0.1)
Annual risk-free rate for risk-adjusted calculations
mi (str): Market index for correlation analysis
Options: "^GSPC" (S&P 500), "^DJI" (Dow Jones), "^IXIC" (NASDAQ), "^RUT" (Russell 2000)
cw (float): Chronos weight in ensemble (0.0-1.0)
Weight for Chronos model in ensemble predictions
tw (float): Technical weight in ensemble (0.0-1.0)
Weight for technical analysis in ensemble predictions
sw (float): Statistical weight in ensemble (0.0-1.0)
Weight for statistical models in ensemble predictions
rrp (int): Number of random real points to include in long-horizon context
usm (bool): Use smoothing
When True, applies smoothing to predictions to reduce noise and improve continuity
smt (str): Smoothing type to use
Options: "exponential", "moving_average", "kalman", "savitzky_golay", "double_exponential", "triple_exponential", "adaptive", "none"
sww (float): Smoothing window size for moving average and Savitzky-Golay
sa (float): Smoothing alpha for exponential methods (0.1-0.9)
Returns:
Tuple[Dict, go.Figure, Dict, Dict, Dict, Dict, Dict, Dict, Dict]: Analysis results containing:
- Dict: Basic trading signals optimized for hourly timeframes
- go.Figure: Interactive plot with hourly data, predictions, and intraday patterns
- Dict: Product metrics including intraday volatility and volume analysis
- Dict: Risk metrics adjusted for hourly data frequency
- Dict: Sector analysis with intraday-specific metrics
- Dict: Market regime detection for hourly patterns
- Dict: Stress testing results for short-term scenarios
- Dict: Ensemble analysis configuration and results
- Dict: Advanced signals with intraday-specific indicators
Raises:
gr.Error: If market is closed, insufficient data, or analysis errors
Hourly data is only available during market hours (9:30 AM - 4:00 PM ET)
Example:
>>> signals, plot, metrics, risk, sector, regime, stress, ensemble, advanced = hourly_analysis(
... "AAPL", 3, 14, "chronos", True, True, True, 0.02, "^GSPC", 0.6, 0.2, 0.2, 4, True, "exponential", 5, 0.3
... )
Notes:
- Only available during market hours (9:30 AM - 4:00 PM ET, weekdays)
- Maximum prediction period is 7 days (168 hours)
- Historical data limited to 60 days due to Yahoo Finance constraints
- Includes pre/post market data for extended hours analysis
- Optimized for day trading and swing trading strategies
- Requires high-liquidity stocks for reliable hourly analysis
- Smoothing helps reduce prediction noise but may reduce responsiveness to sudden changes
"""
return analyze_stock(s, "1h", pd, ld, st, ue, urd, ust, rfr, mi, cw, tw, sw, rrp, usm, smt, sww, sa)
hourly_predict_btn.click(
fn=hourly_analysis,
inputs=[hourly_symbol, hourly_prediction_days, hourly_lookback_days, hourly_strategy,
use_ensemble, use_regime_detection, use_stress_testing, risk_free_rate, market_index,
chronos_weight, technical_weight, statistical_weight,
random_real_points, use_smoothing, smoothing_type, smoothing_window, smoothing_alpha],
outputs=[hourly_signals, hourly_plot, hourly_metrics, hourly_risk_metrics, hourly_sector_metrics,
hourly_regime_metrics, hourly_stress_results, hourly_ensemble_metrics, hourly_signals_advanced]
)
# 15-minute analysis button click
def min15_analysis(s: str, pd: int, ld: int, st: str, ue: bool, urd: bool, ust: bool,
rfr: float, mi: str, cw: float, tw: float, sw: float,
rrp: int, usm: bool, smt: str, sww: float, sa: float) -> Tuple[Dict, go.Figure, Dict, Dict, Dict, Dict, Dict, Dict, Dict]:
"""
Process 15-minute timeframe stock analysis with advanced features.
This function performs ultra-high-frequency stock analysis using 15-minute data,
designed for scalping and very short-term trading strategies. It includes specialized
indicators for intraday patterns, volume analysis, and momentum detection.
Args:
s (str): Stock symbol (e.g., "AAPL", "MSFT", "GOOGL", "TSLA")
Must be a highly liquid stock symbol suitable for high-frequency analysis
pd (int): Number of days to predict (1-2)
Limited to 2 days due to 15-minute data granularity and model constraints
ld (int): Historical lookback period in days (1-7)
Enhanced to 7 days for 15-minute data (vs standard 5 days)
st (str): Prediction strategy to use ("chronos" or "technical")
- "chronos": Uses Amazon's Chronos T5 model optimized for 15-minute intervals
- "technical": Uses technical indicators specifically tuned for 15-minute timeframes
ue (bool): Use ensemble methods
Combines multiple models for improved ultra-short-term prediction accuracy
urd (bool): Use regime detection
Detects micro-market regimes and volatility clustering patterns
ust (bool): Use stress testing
Performs scenario analysis for intraday market shocks and volatility spikes
rfr (float): Risk-free rate (0.0-0.1)
Annual risk-free rate for risk-adjusted calculations (less relevant for 15m analysis)
mi (str): Market index for correlation analysis
Options: "^GSPC" (S&P 500), "^DJI" (Dow Jones), "^IXIC" (NASDAQ), "^RUT" (Russell 2000)
cw (float): Chronos weight in ensemble (0.0-1.0)
Weight for Chronos model in ensemble predictions
tw (float): Technical weight in ensemble (0.0-1.0)
Weight for technical analysis in ensemble predictions
sw (float): Statistical weight in ensemble (0.0-1.0)
Weight for statistical models in ensemble predictions
rrp (int): Number of random real points to include in long-horizon context
usm (bool): Use smoothing
When True, applies smoothing to predictions to reduce noise and improve continuity
smt (str): Smoothing type to use
Options: "exponential", "moving_average", "kalman", "savitzky_golay", "double_exponential", "triple_exponential", "adaptive", "none"
sww (float): Smoothing window size for moving average and Savitzky-Golay
sa (float): Smoothing alpha for exponential methods (0.1-0.9)
Returns:
Tuple[Dict, go.Figure, Dict, Dict, Dict, Dict, Dict, Dict, Dict]: Analysis results containing:
- Dict: Basic trading signals optimized for 15-minute timeframes
- go.Figure: Interactive plot with 15-minute data, predictions, and micro-patterns
- Dict: Product metrics including high-frequency volatility and volume analysis
- Dict: Risk metrics adjusted for 15-minute data frequency
- Dict: Sector analysis with ultra-short-term metrics
- Dict: Market regime detection for 15-minute patterns
- Dict: Stress testing results for intraday scenarios
- Dict: Ensemble analysis configuration and results
- Dict: Advanced signals with 15-minute-specific indicators
Raises:
gr.Error: If market is closed, insufficient data points, or analysis errors
15-minute data requires at least 64 data points and is only available during market hours
Example:
>>> signals, plot, metrics, risk, sector, regime, stress, ensemble, advanced = min15_analysis(
... "AAPL", 1, 3, "chronos", True, True, True, 0.02, "^GSPC", 0.6, 0.2, 0.2, 4, True, "exponential", 5, 0.3
... )
Notes:
- Only available during market hours (9:30 AM - 4:00 PM ET, weekdays)
- Maximum prediction period is 2 days (192 15-minute intervals)
- Historical data limited to 7 days due to Yahoo Finance constraints
- Requires minimum 64 data points for reliable Chronos predictions
- Optimized for scalping and very short-term trading strategies
- Includes specialized indicators for intraday momentum and volume analysis
- Higher transaction costs and slippage considerations for 15-minute strategies
- Best suited for highly liquid large-cap stocks with tight bid-ask spreads
- Smoothing helps reduce prediction noise but may reduce responsiveness to sudden changes
"""
return analyze_stock(s, "15m", pd, ld, st, ue, urd, ust, rfr, mi, cw, tw, sw, rrp, usm, smt, sww, sa)
min15_predict_btn.click(
fn=min15_analysis,
inputs=[min15_symbol, min15_prediction_days, min15_lookback_days, min15_strategy,
use_ensemble, use_regime_detection, use_stress_testing, risk_free_rate, market_index,
chronos_weight, technical_weight, statistical_weight,
random_real_points, use_smoothing, smoothing_type, smoothing_window, smoothing_alpha],
outputs=[min15_signals, min15_plot, min15_metrics, min15_risk_metrics, min15_sector_metrics,
min15_regime_metrics, min15_stress_results, min15_ensemble_metrics, min15_signals_advanced]
)
return demo
if __name__ == "__main__":
demo = create_interface()
demo.launch(ssr_mode=False, mcp_server=True)