TRADING BOT

A MINI PROJECT-I REPORT

Submitted by

MURAJIT VARUN K (2211039)

GOWTHAM S (2211021)

in partial fulfillment for the award of the degree

of

BACHELOR OF TECHNOLOGY

in

ARTIFICIAL INTELLIGENCE AND DATA SCIENCE

SRI RAMAKRISHNA ENGINEERING COLLEGE

[Educational Service: SNR Sons Charitable Trust]
[Autonomous Institution, Reaccredited by NAAC with‘A+’ Grade]
[Approved by AICTE and Permanently Affiliated to Anna University, Chennai]
[ISO 9001:2015 Certified and All Eligible Programmes Accredited by NBA]
Vattamalaipalayam, N.G.G.O. Colony Post,

COIMBATORE – 641 022

ANNA UNIVERSITY : CHENNAI 600 025

MAY 2024
 TRADING BOT
A MINI PROJECT-I REPORT

Submitted by

MURAJIT VARUN K (2211039)

GOWTHAM S (2211021)

in partial fulfillment for the award of the degree

of

BACHELOR OF TECHNOLOGY

in

ARTIFICIAL INTELLIGENCE AND DATA SCIENCE

SRI RAMAKRISHNA ENGINEERING COLLEGE

[Educational Service: SNR Sons Charitable Trust]
[Autonomous Institution, Reaccredited by NAAC with‘A+’ Grade]
[Approved by AICTE and Permanently Affiliated to Anna University, Chennai]
[ISO 9001:2015 Certified and All Eligible Programmes Accredited by NBA]
Vattamalaipalayam, N.G.G.O. Colony Post,

COIMBATORE – 641 022

ANNA UNIVERSITY: CHENNAI 600 025

MAY 2024

ii
 ANNA UNIVERSITY: CHENNAI 600 025

BONAFIDE CERTIFICATE

20AD272 - MINI PROJECT I

Certified that this project Report “TRADING BOT” is the bonafide work of “MURAJIT
VARUN(22110139), GOWTHAMS(2211021)” who carried ou22222t the project work
under my supervision.

SIGNATURE SIGNATURE
Dr.V.Karpagam Mrs.C.Kavitha
HEAD OF THE DEPARTMENT SUPERVISOR
Professor, Assistant Professor,
Department of Artificial Intelligence and Department of Artificial Intelligence and
Data Science, Data Science,
Sri Ramakrishna Engineering College, Sri Ramakrishna Engineering College,
Coimbatore-641022. Coimbatore-641022.

Submitted for the Mini Project Viva-Voce examination held on ______________

INTERNAL EXAMINER EXTERNAL EXAMINER

iii
 ACKNOWLEDGEMENT

We thank the almighty for his blessings on us to complete this project

work successfully.
With a profound sense of gratitude, we thank the Management, Managing
Trustee, Thiru D.Lakshminarayanasamy, and Joint Managing Trustee, Thiru
R. Sundar, for having provided us with the necessary infrastructure required for
the completion of our project.

With a profound sense of gratitude, we sincerely thank the Head of the

Institution, Dr. N. R. Alamelu, for her kind patronage, which helped in pursuing
the project successfully.

With immense pleasure, we express our hearty thanks to the Head of the
Department, Dr. V. Karpagam, Department of Artificial Intelligence and Data
Science for her encouragement towards the completion of this project.

We thank our Academic Coordinator Mr M. Logaprakash Assistant

Professor and Project Coordinator Mr M. Logaprakash, Assistant Professor,
Department of Information Technology for his encouragement towards the
completion of this project.

We also thank our Project guide Mrs C. Kavitha Asst Professor,

Department of Artificial Intelligence & Data Science for her guidance towards
the completion of the project.

We convey our thanks to all the teaching and non-teaching staff members
of our department who rendered their co-operation by all means for
completion of this project.
iv
 ABSTRACT

This project presents a trading strategy for financial markets based on sentiment
analysis, using natural language processing (NLP) techniques and machine learning models.
The strategy uses sentiment analysis of newspaper headlines to decide whether to buy or sell
a given financial instrument. The implementation uses the Lumibot framework for broker
integration and backtesting along with the Alpaca API for accessing financial data and
executing trades.
The sentiment analysis component uses the FinBERT model, a pre-trained
transformer-based model tuned for financial sentiment analysis. News headlines are
processed using the FinBERT model to predict sentiment (positive, negative or neutral) and
associated probabilities. The trading strategy dynamically adjusts position size and trade
execution based on the results of sentiment analysis in order to take advantage of sentiment-
driven market movements.
The script includes functions to estimate sentiment for a given set of news headlines,
check GPU availability for accelerated computation using CUDA, and backtest a trading
strategy based on sentiment analysis using historical financial data. Backtest evaluates the
performance of a trading strategy over a certain period of time and provides insight into its
profitability and effectiveness. Overall, this project demonstrates the potential of sentiment
analysis techniques in developing algorithmic trading strategies for finance
markets that offer opportunities for automated decision making and improved risk
management.

v
 TABLE OF CONTENTS

Chapter No. Title Page No.

ABSTRACT iv
LIST OF ABBREVIATIONS vii

1 INTRODUCTION 1
PROBLEM STATEMENT 2
OBJECTIVE 2
2 LITERATURE SURVEY 3
3 3.1 EXISTING SYSTEMS 5

3.2 PROPOSED SYSTEM 5

3.3 SENTIMENT ANALYSIS ON 6

FINANCIAL
3.4 ALGORITHMIC TRADING 7
STRATEGIES
3.5NLP AND MACHINELEARNING 8
MODELS

4 WORKING PRINCIPLE 9

4.1 DATA COLLECTION 9

vi
 4.2 DATA PREPROCESSING 10

4.3 SENTIMENT ANALYSIS

5 5.1ALGORITHMIC TRADING STRATEGIES 12

5.2 BACKTESTING FRAMEWORK 13

5.3 MODULES 14

6 CONCLUSION AND FUTURE SCOPE 15

APPENDIXES

APPENDIX 1 15
APPENDIX 2 18

SCREEN SHOTS 23

REFERENCES 26

vii
 LIST OF ABBREVIATIONS

ABBREVIATION EXPANSION

Application Programming
API
Interface
Representational State
REST
Transfer
API_KEY
Application Programming
Interface Key
Application Programming
API_SECRET
Interface Secret
SQL Structured Query Language

EM Expectation-Maximization

PT PyTorch

NLP Natural Language Processing

viii
 CHAPTER 1
INTRODUCTION

In today's fast-moving financial markets, the use of cutting-edge technology has become
indispensable for traders looking for an edge. Enter the world of AI trading bots – sophisticated
algorithms designed to analyze market data, identify patterns and execute trades with accuracy
and speed. Powered by artificial intelligence and machine learning, these bots have
revolutionized the way investors navigate the complexities of trading.
At its core, an AI trading bot is a culmination of advanced mathematical models and
computing power, capable of processing massive amounts of data in real time. Using historical
market data, news sentiment analysis and technical indicators, these bots can identify trading
opportunities that human traders might miss. Through iterative learning processes, they
constantly adapt and optimize their strategies to outperform traditional methods and generate
consistent returns.
The benefits of AI trading bots are numerous. They operate 24 hours a day, 7 days a
week, unaffected by human emotion or fatigue, and ensure continuous vigilance over market
movements. Their lightning-fast execution capabilities allow them to take advantage of
fleeting opportunities and respond immediately to market changes. Additionally, their ability
to analyze multiple variables simultaneously gives them a holistic view of market dynamics,
allowing them to make informed decisions with greater accuracy.
However, it is essential to realize that AI trading robots are not infallible. Market
conditions can be unpredictable and even the most sophisticated algorithms can run into
unexpected problems. In addition, regulatory scrutiny and ethical considerations regarding
algorithmic trading underscore the importance of responsible use and oversight.

i
 PROBLEM STATEMENT

Effective use of market sentiment to inform trading decisions remains a significant

challenge in the field of algorithmic trading. Despite the availability of vast amounts of text
data, including financial news headlines and social media posts, marketers often struggle to
extract actionable insights from this unstructured data. Traditional market analysis methods
such as technical analysis and fundamental analysis have limitations in capturing the nuances
expressed by market participants.
The problem statement that this trading robot program solves is twofold:
Interpretation of market sentiment:
Traders face difficulties in accurately interpreting market sentiment from textual data
due to the complexity of language, changing sentiments and noise in the data.
Identifying trends, sentiments and emotions expressed in financial news headlines and
social media posts is challenging without the help of automated sentiment analysis techniques.
OBJECTIVE

Develop a trading strategy based on sentiment analysis:

Design a trading strategy that incorporates sentiment analysis techniques to make informed
trading decisions.
Implement machine learning techniques:
Deploy machine learning models to automate sentiment analysis and trading decisions.
Evaluate the performance of the trading strategy:
Assess trading strategy performance through rigorous back-testing and analysis.
Send merchants with advanced tools:

1
 Give marketers a tool that improves decision making by leveraging sentiment analysis
insights.

CHAPTER 2

LITERATURE SURVEY

2.1 Jerry Joy, Aparna Kannan, Shreya Ram,“ Speech emotion recognition using
neural network and MLP classifier” ISSN 2321 3361 © 2020 IJESC
It is suggested to use a deep learning model that has had its hyperparameters
optimized to find the ideal settings and get more accurate recognition results. This deep
learning model consists of a convolutional neural network (CNN) layer and four local
feature-learning blocks to learn short- and long-term correlations in the log Mel-spectrogram
of the input speech samples. To show the effectiveness of the method, four speech emotion
datasets—IEMOCAP, Emo-DB, RAVDESS, and SAVEE—are included in the experiments.
Based on the four datasets, the attained recognition accuracies are 98.13%, 99.76%, 99.47%,
and 99.50%, respectively.

2.2 Felicia Andayani, Lau Bee Theng Mark Teekit Tsun and CasloChua “Hybrid
LSTM-Transformer Model for Emotion Recognition from Speech Audio Files” IEEE
Access.2022.3163856
This research introduced a hybrid Long Short-Term Memory (LSTM) Network and
Transformer Encoder to analyze the long-term relationships in voice signals and discern
emotions. The Mel Frequency Cepstral Coefficient (MFCC) is used to extract speech
features, which are then fed into the proposed hybrid LSTM-Transformer classifier. Several
performance evaluations of the proposed LSTM-Transformer model were conducted. The
results demonstrate that, in terms of recognition, it greatly outperforms the models currently
offered by earlier published studies. The recommended hybrid model achieved recognition

2
 success rates of 75.62%, 85.55%, and 72.49%, respectively, using the RAVDESS, EmoDB,
and language-independent datasets.

2.3 Taiba majid wani, Teddy surya gunawan ,Syed asif ahma “A comprehensive Review
of Speech Emotion and Recognition Systems” IEEE-EXPLORE 9383000
In the literature on speech emotion recognition, numerous techniques, including
numerous well-known speech analysis and classification techniques, have been employed to
extract emotions from signals. (SER). Deep learning methodologies have recently been
presented as a replacement for traditional SER procedures. Some of the acoustic features
used are MFCCs, PLPs, and FBANKs. Combining these three acoustic features with MFCCs
resulted in better performance and a higher identification rate of 92.3% than when utilizing
only one acoustic feature, which was 92.1%.

2.4 Nagaraja N Poojary, Dr.Shivakumar.G.S, Akshath Kumar. B.H "Speech Emotion

Recognition Using MLP Classifier" DOI:10.32628/CSEIT217446
Emotions play an important role in social interactions. Since we are dealing with
human-machine interaction, identifying emotion in speech is both very important and
difficult. The goal of our project is to generate intelligent speech that recognizes emotions
using convolutional neural networks. Several modules are used for emotion detection and
classifiers are used to distinguish emotions such as happiness, sadness, anger, and surprise.
The data are audio samples and the librosa package is used to extract attributes from the
audio samples. As an experimental dataset, we use the RAVDESS dataset. The study shows
an accuracy of 68% for all classifiers in the dataset.

3
 CHAPTER 3
3.1 EXISTING SYSTEMS

QuantConnect is a cloud-based algorithmic trading platform that allows users to design,

backtest, and deploy trading algorithms across multiple asset classes. It supports various
programming languages such as C#, Python, and F#, and provides access to historical data,
fundamental data, and alternative data sources.

TradeStation is a brokerage platform that offers a suite of trading tools and analytics
for active traders. It provides a platform called EasyLanguage for developing custom trading
strategies and indicators. TradeStation also offers a marketplace where users can purchase
or sell trading strategies developed by other users

3.2 PROPOSED SYSTEM

Integrating sentiment analysis into business strategies:

The proposed system aims to integrate sentiment analysis techniques into algorithmic trading
strategies to improve decision making and trading performance.

Using sentiment analysis from textual data sources such as financial news headlines and
social media posts, the system aims to identify sentiment-driven market trends and capitalize
on trading opportunities.

Automated sentiment analysis:

The system will include an automated sentiment analysis pipeline to preprocess text data,
extract sentiment features, and classify text into sentiment categories.

4
 3.3 SENTIMENT ANALYSIS ON FINANCIAL

1. Importance of sentiment analysis:

• Sentiment analysis plays a key role in financial markets by providing insight into investor
sentiment, market sentiment and economic trends.
• Understanding sentiment enables traders and investors to gauge market sentiment, predict
market movements and make informed trading decisions.
2. Sentiment Data Sources:
• Financial market sentiment data can be derived from a variety of sources, including
financial news articles, social media platforms (eg Twitter, StockTwits), earnings transcripts
and analyst reports.
• These sources provide textual data that can be analyzed for sentiment and sentiment-related
information.
3. Sentiment analysis methods:
• Sentiment analysis methods in financial markets range from rule-based approaches to
machine learning and natural language processing (NLP) techniques.
• Rule-based approaches use predefined rules and lexicons to categorize text as positive,
negative, or neutral based on sentiment indicators.
• Machine learning and NLP techniques involve training models on labeled data to classify
text based on sentiment, enabling more detailed analysis and interpretation.
4. Application of sentiment analysis:
• Sentiment analysis is used in various areas of financial markets, including stock market
prediction, asset allocation, risk management, and sentiment-driven trading strategies.
• Traders and investors use sentiment analysis to identify sentiment-driven market trends,
sentiment anomalies, and sentiment-driven price movements.

5
 3.4 ALGORITHMIC TRADING STRATEGIES

Definition and overview:

Algorithmic trading strategies are automated trading strategies that use computer algorithms
to execute trades in the financial markets.
The goal of these strategies is to exploit market inefficiencies, price discrepancies and
trading opportunities using quantitative analysis and mathematical models.
Types of Algorithmic Trading Strategies:
There are different types of algorithmic trading strategies, including:
Momentum Strategy: Trading based on the momentum of price movements.
Mean reversion strategy: Trading based on the tendency of prices to revert to their mean
or average value.
Arbitrage Strategy: Exploiting price differences between related assets or markets.
Market Making Strategy: Providing liquidity by constantly quoting bid and ask prices.
Sentiment Based Strategies: Trading based on market sentiment derived from sentiment
analysis of newspaper headlines and social media data.
Advantages of algorithmic trading strategies:
Algorithmic trading strategies offer several advantages over manual trading, including:
Speed: Algorithms can execute trades much faster than human traders, allowing for rapid
response to market events and opportunities.
Efficiency: Algorithms can analyze large data sets and execute trades with accuracy and
consistency, reducing the impact of human emotions and biases.
Scalability: Algorithmic trading strategies can be scaled to trade multiple markets and assets
simultaneously, increasing trading volume and liquidity.
Challenges and risks:
Despite its advantages, algorithmic trading strategies also face challenges and risks,
including:

6
 3.5 NLP AND MACHINE LEARNING MODELS
Role in Sentiment Analysis:

Natural language processing (NLP) and machine learning models play a key role in sentiment
analysis by enabling automated sentiment extraction and analysis from textual data.

NLP techniques are used to preprocess text data, tokenize words, remove stop words, and
extract features for sentiment analysis.

Machine learning models are trained on labeled data to classify text as positive, negative, or
neutral based on sentiment indicators, enabling automated sentiment analysis.

Types of Machine Learning Models:

Various machine learning models are used in sentiment analysis, including:

Supervised learning models: Such as Support Vector Machines (SVM), Naive Bayes, and
Logistic Regression, which learn to classify text based on labeled training data.

Deep learning models: such as recurrent neural networks (RNNs), convolutional neural
networks (CNNs), and transformer models that learn to capture complex patterns and
relationships in text data for sentiment analysis.

Ensemble methods: Such as Random Forests and Gradient Boosting Machines (GBM),
which combine multiple underlying models to improve prediction accuracy.

Function:

The engineering function plays a key role in training machine learning models for sentiment
analysis.

Features such as word embeddings, n-grams, parts of speech, and syntactic or semantic
features are extracted from text data to capture sentiment-related information.

7
 CHAPTER 4

WORKING PRINCIPLE

4.1 DATA COLLECTION

The methodology begins by identifying relevant data sources for market data collection and
textual data for sentiment analysis.
Market data sources may include financial APIs, market exchanges, and data vendors, while
text data sources may include financial news websites, social media platforms, and analyst
reports.
PI integration and data acuisition:
The methodology includes integration with various APIs to retrieve market data, including
historical price data, trade volumes and other relevant market indicators.
APIs provided by financial data providers and market exchanges are used to access real-time
market data streams and historical data streams.
Web scraping and text data extraction:
Web scraping techniques are used to extract textual data from online sources such as financial
news websites and social media platforms.
Text data extraction involves parsing HTML content, identifying relevant text elements, and
extracting news headlines, article content, and social media posts.
Data preprocessing:
Pre-processing steps are applied to both market data and text data to prepare it for analysis.
Market data pre-processing can include cleaning and filtering noise, handling missing values
and standardizing data formats.
Text data preprocessing can include text cleaning, tokenization, lemmatization, and stop word
removal to prepare the text for sentiment analysis.
Data storage and management:
Collected and pre-processed data is stored in a structured format for easy access and analysis.
8
 4.2 DATA PREPROCESSING

1. Text cleaning:
• Text cleaning involves removing unnecessary characters, symbols and formatting from text
data.
• Common sanitization techniques include removing special characters, punctuation, HTML
and URL tags from text.
2. Tokenization:
• Tokenization is the process of dividing text into individual words or tokens.
• Text is tokenized based on spaces or punctuation, creating a list of words or tokens for further
analysis.
3. Normalization:
• Normalization techniques are used to standardize textual data and reduce variation.
• This may include converting text to lowercase, removing diacritics or diacritics, and
expanding contractions (eg converting "can't" to "can't").
4. Stopword removal:
• Stopwords are common words that do not have a significant meaning when analyzing the text.
• Ignored word removal involves filtering out the ignored words from the text to focus on the
meaningful content.
• Common ignored words include articles, conjunctions, and prepositions.
5. Lemmatization and stemming:
• Lemmatization and stemming are techniques used to reduce words to their root form.
• Lemmatization maps words to their base or dictionary form (lemma), while lemma removes
prefixes and suffixes to form the root or stem of a word.

9
4.3 SENTIMENT ANALYSIS
Definition:
Sentiment analysis, also known as opinion mining, is the process of computationally
identifying and categorizing opinions, attitudes, and emotions expressed in textual data.
Objective:
The primary goal of sentiment analysis is to determine the sentiment polarity of a text and
classify it as positive, negative, or neutral based on the sentiment expressed.
Application:
Sentiment analysis has various applications in different domains, including:
Marketing: Analyzing customer feedback, reviews and social media posts to understand
customer sentiment towards products and brands.
Finance: Analyzing financial news headlines, earnings call transcripts and social media
discussions to gauge market sentiment and make investment decisions.
Customer Service: Monitoring customer feedback and sentiment on social media platforms
to identify issues, trends and opportunities for improvement.
Politics: Analysis of public opinion and sentiment toward political candidates, parties, and
policies during elections or campaigns.
Techniques:
Sentiment analysis techniques can generally be divided into:
Rule-based approaches: Using predefined rules and lexicons to identify sentiment
indicators and classify text based on predefined criteria.
Machine learning approaches: Train machine learning models on labeled data to
automatically learn and classify text based on sentiment features.
Hybrid approaches: Combining rule-based and machine learning techniques to leverage
the strengths of both approaches for more accurate sentiment analysis.

10
 CHAPTER 5
5.1ALGORITHMIC TRADING STRATEGIES
Definition and overview:
Algorithmic trading strategies are automated trading strategies that use computer algorithms
to execute trades in the financial markets.
The goal of these strategies is to exploit market inefficiencies, price discrepancies and
trading opportunities using quantitative analysis and mathematical models.
Types of Algorithmic Trading Strategies:
There are different types of algorithmic trading strategies, including:
Momentum Strategy: Trading based on the momentum of price movements.
Mean reversion strategy: Trading based on the tendency of prices to revert to their mean
or average value.
Arbitrage Strategy: Exploiting price differences between related assets or markets.
Market Making Strategy: Providing liquidity by constantly quoting bid and ask prices.
Sentiment Based Strategies: Trading based on market sentiment derived from sentiment
analysis of newspaper headlines and social media data.
Advantages of algorithmic trading strategies:
Algorithmic trading strategies offer several advantages over manual trading, including:
Speed: Algorithms can execute trades much faster than human traders, allowing for rapid
response to market events and opportunities.
Efficiency: Algorithms can analyze large data sets and execute trades with accuracy and
consistency, reducing the impact of human emotions and biases.
Scalability: Algorithmic trading strategies can be scaled to trade multiple markets and assets
simultaneously, increasing trading volume and liquidity.

11
 5.2 BACKTESTING FRAMEWORK
Definition:
A backtesting framework is a systematic approach used to evaluate the performance of
trading strategies using historical market data.
It includes the simulation of trade execution based on predefined rules and parameters to
assess the effectiveness and profitability of the strategy.
Purpose:
The primary purpose of the backtesting framework is to validate trading strategies and
evaluate their performance under historical market conditions.
It helps traders identify strengths and weaknesses and areas for improvement in their
strategies before deploying them in a live trading environment.
Backtesting Framework components:
A backtesting framework typically consists of the following components:
Historical Data: Market data over a period of time used for backtesting.
Strategy Implementation: The code or algorithms that define the rules, conditions, and
parameters of a trading strategy.
Simulation Engine: A software or platform that executes trades based on trading strategy
and historical data.
Performance Metrics: Metrics used to evaluate strategy performance, such as profitability,
risk-adjusted returns, drawdowns, and Sharpe ratio.
Visualization and Reporting: Tools to visualize backtesting results and generate
performance reports to analyze and interpret strategy performance.
Backtesting Process:
The backtesting process includes the following steps:
Data Preparation: Cleaning, pre-processing and formatting historical market data for
analysis.

12
 5.3 MODULES
Sentiment Analysis Module:
The sentiment analysis module is responsible for analyzing textual data from sources such
as financial news headlines and social media posts for sentiment indicators.
Natural language processing (NLP) techniques and machine learning models are used to
classify text into positive, negative, or neutral sentiment categories.
The module preprocesses text data, tokenizes words, removes ignored words, and applies
sentiment analysis algorithms to generate sentiment scores or labels.
Business Strategy Module:
The trading strategy module designs and implements algorithmic trading strategies that
incorporate insights from sentiment analysis.
It defines the rules, conditions and parameters for making trading decisions based on the
outputs of sentiment analysis, market data and risk management principles.
Machine learning models can be used to develop predictive trading signals and optimize
strategy performance.
Backtesting module:
The Backtesting module evaluates the performance of trading strategies using historical
market data.
It simulates the execution of trades based on pre-defined strategic rules and parameters to
assess profitability, risk and robustness.
Performance metrics such as Sharpe ratio, maximum draw and win-loss ratio are calculated
to analyze the strategy's performance under different market conditions.
Data collection module:
The data collection module acquires and pre-processes the market data and text data needed
for sentiment analysis and backtesting.
Integrates with financial APIs, market exchanges and data vendors to access real-time and
historical market data streams.
13
 CHAPTER-6

CONCLUSION AND FUTURE SCOPE

6.1 Conclusion:
The proposed system demonstrates the feasibility and effectiveness of integrating sentiment
analysis techniques into algorithmic trading strategies.
Through backtesting and performance evaluation, the system has shown promising results in
generating profits and managing risks in the financial markets.
Using sentiment analysis provides valuable insights into market sentiment and helps traders
make informed decisions based on sentiment-driven trends.
6.2 Future improvements:
Several avenues for future enhancements and improvements have been identified:
Integrating Alternative Data Sources: Incorporating additional data sources such as alternative
data, social media sentiment, and macroeconomic indicators to increase predictive power and
robustness.
Advanced Machine Learning Techniques: Exploring advanced machine learning models
such as deep learning architectures and reinforcement learning algorithms for more accurate
sentiment analysis and predictive modeling.
Real-Time Sentiment Analysis: Developing real-time sentiment analysis capabilities to
capture dynamic changes in market sentiment and adjust trading strategies accordingly.
Dynamic Strategy Optimization: Implementation of dynamic strategy optimization
techniques that adjust strategy parameters in response to evolving market conditions and
sentiment analysis.
Compliance and risk management: Strengthening compliance measures and risk
management protocols to ensure compliance with regulatory requirements and mitigate
potential legal and financial risks.

14
 APPENDIX 1

User interface
import tkinter as tk
from tkinter import ttk

class TradingBotUI:
def __init__(self, master):
self.master = master
master.title("Trading Bot")

self.main_frame = ttk.Frame(master, padding="20")

self.main_frame.grid(row=0, column=0)

self.label = ttk.Label(self.main_frame, text="Enter your trade parameters:",

font=("Arial", 14))
self.label.grid(row=0, column=0, columnspan=2, pady=10)

self.apikey_label = ttk.Label(self.main_frame, text="API Key:", font=("Arial", 12))

self.apikey_label.grid(row=1, column=0, padx=10, pady=5, sticky="e")

self.apikey_entry = ttk.Entry(self.main_frame, font=("Arial", 12))

self.apikey_entry.grid(row=1, column=1, padx=10, pady=5, sticky="w")

self.apisecret_label = ttk.Label(self.main_frame, text="API Secret:", font=("Arial",

12))
self.apisecret_label.grid(row=2, column=0, padx=10, pady=5, sticky="e")

15
 self.apisecret_entry = ttk.Entry(self.main_frame, font=("Arial", 12))
self.apisecret_entry.grid(row=2, column=1, padx=10, pady=5, sticky="w")

self.submit_button = ttk.Button(self.main_frame, text="Submit",

command=self.submit)
self.submit_button.grid(row=3, column=0, columnspan=2, pady=10)

def submit(self):
apikey = self.apikey_entry.get()
apisecret = self.apisecret_entry.get()

# Here you would call your trading bot function with the provided parameters
# Example:
# trading_bot(apikey, apisecret)

# For now, let's just print the inputs

print("API Key:", apikey)
print("API Secret:", apisecret)

def main():
root = tk.Tk()
root.geometry("300x200")
trading_bot_ui = TradingBotUI(root)
root.mainloop()

if __name__ == "__main__":
main()

16
 APPENDIX 2

Trading bot:

import lumibot
import datetime
import alpaca_trade_api
from lumibot.backtesting import YahooDataBacktesting
from lumibot.strategies.strategy import Strategy
from lumibot.traders import Trader
from datetime import datetime
from alpaca_trade_api import REST
from timedelta import Timedelta
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
from typing import Tuple
device ="cuda:0" if torch.cuda.is_available() else "cpu"

API_KEY = "PKBFY9NDMDZ3DZGNUBHU"
API_SECRET = "Q8Y6j2oTqTML1cOaXIHZSXbYYszZJtxZsZLrIT0p"
BASE_URL = "https://paper-api.alpaca.markets"

ALPACA_CREDS = {
"API_KEY":API_KEY,
"API_SECRET": API_SECRET,
"PAPER": True
}

17
tokenizer = AutoTokenizer.from_pretrained("ProsusAI/finbert")
model =
AutoModelForSequenceClassification.from_pretrained("ProsusAI/finbert").to(device)
labels = ["positive", "negative", "neutral"]

def estimate_sentiment(news):
if news:
tokens = tokenizer(news, return_tensors="pt", padding=True).to(device)

result = model(tokens["input_ids"], attention_mask=tokens["attention_mask"])[

"logits"
]
result = torch.nn.functional.softmax(torch.sum(result, 0), dim=-1)
probability = result[torch.argmax(result)]
sentiment = labels[torch.argmax(result)]
return probability, sentiment
else:
return 0, labels[-1]

class MLTrader(Strategy):
def initialize(self, symbol:str="SPY", cash_at_risk:float=.5):
self.symbol = symbol
self.sleeptime = "24H"
self.last_trade = None
self.cash_at_risk = cash_at_risk
self.api = REST(base_url=BASE_URL,key_id=API_KEY, secret_key=API_SECRET)

18
def position_sizing(self):
cash = self.get_cash()
last_price = self.get_last_price(self.symbol)
quantity = round(cash * self.cash_at_risk / last_price,0)
return cash, last_price, quantity

def get_dates(self):
today = self.get_datetime()
three_days_prior = today - Timedelta(days=3)
return today.strftime('%Y-%m-%d'), three_days_prior.strftime('%Y-%m-%d')

def get_sentiment(self):
today, three_days_prior = self.get_dates()
news = self.api.get_news(symbol=self.symbol,
start=three_days_prior,
end=today)
news = [ev.__dict__["_raw"]["headline"] for ev in news]
probability, sentiment = estimate_sentiment(news)
return probability, sentiment

def on_trading_iteration(self):
cash, last_price, quantity = self.position_sizing()
probability, sentiment = self.get_sentiment()

if cash > last_price:

if sentiment == "positive" and probability > .999:
if self.last_trade == "sell":

19
 self.sell_all()
order = self.create_order(
self.symbol,
quantity,
"buy",
type="bracket",
take_profit_price=last_price*1.20,
stop_loss_price=last_price*.95
)
self.submit_order(order)
self.last_trade = "buy"
elif sentiment == "negative" and probability > .999:
if self.last_trade == "buy":
self.sell_all()
order = self.create_order(
self.symbol,
quantity,
"sell",
type="bracket",
take_profit_price=last_price*.8,
stop_loss_price=last_price*1.05
)
self.submit_order(order)
self.last_trade = "sell"

start_date = datetime(2020,1,1)
end_date = datetime(2023,12,31)

20
broker = alpaca(ALPACA_CREDS)
strategy = MLTrader(name='mlstrat', broker=broker,
parameters={"symbol":"SPY",
"cash_at_risk":.5})
strategy.backtest(
YahooDataBacktesting,
start_date,
end_date,
parameters={"symbol":"SPY", "cash_at_risk":.5}
)
# trader = Trader()
# trader.add_strategy(strategy)
# trader.run_all()

21
SCREEN SHOTS

22
23
24
 REFERENCES

1. Cambria, E., Schuller, B., Xia, Y., & Havasi, C. (2013). New avenues in opinion
gathering and sentiment analysis. IEEE Intelligent Systems, 28(2), 15–21.
https://doi.org/10.1109/MIS.2013.30

2. Hull, J.C. (2014). Options, Futures, and Other Derivatives (9th ed.). Pearson
Education Limited.

3. Kumar, V., & Goyal, D. (2015). Sentiment Analysis: An Overview of the

Development of Advanced Techniques. International Journal of Computer
Applications, 114(7), 29–34. https://doi.org/10.5120/20174-3650

4. Lohiya, N., Jain, V., & Varshney, D. (2016). A Survey on Sentiment Analysis and
Opinion Mining Techniques. International Journal of Computer Applications, 139(3),
1–5. https://doi.org/10.5120/ijca2016908509

5. Malkiel, B.G., & Fama, E.F. (2014). The Wall Street Random Walk: A Time-Tested
Strategy for Successful Investing (11th ed.). W. W. Norton & Company.

25