Stock Markets Analytics Zoomcamp 2025

Homework 4: Trading Simulation Statistics

Distribution of scores and reported study time for this homework.

Submissions

32

Median total score

14

Average total score

12

Score distribution

All values are points.

Questions score

Min
2
Median
14.0
Max
14
Q1
10.0
Avg
11.8
Q3
14.0

Learning in public score

Min
-
Median
0.0
Max
3
Q1
0.0
Avg
0.3
Q3
0.0

Total score

Min
2
Median
14.0
Max
17
Q1
10.0
Avg
12.0
Q3
14.0

Time distribution

All values are hours reported by students.

Lectures

Min
1.0
Median
2.0
Max
20.0
Q1
2.0
Avg
4.6
Q3
4.0

Homework

Min
1.0
Median
4.5
Max
20.0
Q1
2.3
Avg
5.6
Q3
6.0

Question breakdown

Correctness and answer distribution per question.

1. Find the CAGR for the pred9_clf_second_best_rule_92 using the new setup (with fewer dummy variables)

28 / 32 correct (87.5%)

1 1.1571 2 (6.2%)
2 1.1594 28 (87.5%)
3 1.1647 2 (6.2%)
4 1.1685 0 (0.0%)

2. What is the probability threshold for the best Random Forest decision rule?

24 / 32 correct (75.0%)

1 0.58 0 (0.0%)
2 0.60 4 (12.5%)
3 0.62 3 (9.4%)
4 0.64 24 (75.0%)

3. What is the CAGR for the best predictor of strong future growth (after redefining the target variable)?

26 / 32 correct (81.2%)

1 1.1212 26 (81.2%)
2 1.1313 1 (3.1%)
3 1.1414 3 (9.4%)
4 1.1515 2 (6.2%)

4. [EXPLORATORY] Describe Your Ideal Strategy

32 / 32 correct (100.0%)

Answer Count
Take the best predictions above the threshold 0.7 and invest capital proportionally to the prediction confidence 1
I will have a weighted rule with stop loss/gain thresholds, i.e. if there is consecutive positive predictions then increase the weight of investment upto a limit. 1
Each day, select up to 3 assets with the highest prediction probabilities above 0.7 (i.e., strong confidence), and invest an equal share of capital into those. Do not invest in any asset if its prediction probability is below 0.7. 1
My strategy uses a two-tier approach: Primary investments in Random Forest predictions (threshold ≥0.65) for high-confidence trades Secondary investments in Decision Tree (pred5_clf_10) predictions for diversification I allocate 60% of daily capital to RF trades and 40% to DT, with: Maximum 5 trades/day 10% stop-loss per position Equal weighting across selected trades 1
I want to have less trades with bigger capital as my bank charges 3€ no matter the amount of money I invest. So I will rather invest only in >90% probability. I also want to save time so I would like to find out how can I limit the days when I perform operations, if statistically it is better to buy on some weekday or particular week of the month. 1
✅ My Ideal Trading Strategy My ideal strategy focuses on precision, capital efficiency, and risk control: 📊 Model: Use the most precise rule from the Random Forest (threshold ≥ 0.63), limited to 5–7 trades/day, for high-quality signals. 💰 Capital Allocation: Allocate capital proportionally to prediction confidence (e.g., higher confidence → higher bet size). 🧱 Risk Management: Introduce a cool-down period to avoid re-entering the same asset too quickly. 🔁 Diversification: Enforce sector & region limits to reduce concentration risk. 🔍 Monitoring: Backtest continuously using rolling windows & monitor live stats (precision, trade frequency, drawdown). ⚙️ Execution: Ideally deploy via an MLOps pipeline with trigger-based rebalancing. 🎯 Objective: Keep the system simple, interpretable, and robust to regime changes, while maintaining CAGR ≥ 10% and avoiding overtrading. 1
Since my capital is limited and my strategy is not purely theoretical, I might try to invest $50 in an index fund like the S&P and $100 in a portfolio of stocks that I would predict to be tariff proof. Instead of 30 days, I would make my predictions for at least 3 months and possibly longer and would make the threshhold for a positive prediction higher to limit transaction fees. I might vary the 1/3 ratio of passive to active to make more predictions. The best predictions would require less than 5 trades per year and with a CAGR >= 1.12. I might have to adjust my portfolio to obtain this. I would need to try various ML models as well. 1
invest only with high probability of return, with target with higher treashold: do smaller trading deals but with the biggest certainty. Take simple base rules (like tree), not complex model. 1
My ideal strategy balances precision, diversification, and risk management, while reducing excessive trades. I would use predictions from the best-performing model (e.g., pred5_clf_10) but only act when the model assigns a high probability (≥ 0.70) to a positive return. This helps filter out weak signals. On each trading day, I would: Select the top 3–5 predictions with the highest probabilities above the 0.70 threshold. Allocate funds equally across them (e.g., $100 each). Skip the day entirely if there are no strong signals (all probabilities below 0.70). To manage risk, I’d: Cap daily investment at $500 to avoid overexposure. Monitor drawdowns and suspend trading if cumulative losses exceed a certain threshold (e.g., 10%). Over time, I would analyze performance by sector and gradually exclude underperforming tickers from future predictions, improving returns through data-driven refinement. 1
My Ideal Trading Strategy: I would combine predictions from multiple models to improve robustness. Specifically, I would select the top 5 stocks for which all models agree on a strong upward trend, using a high probability threshold (e.g., >0.75). To reduce risk and avoid overtrading, I would limit my investments to only those stocks that surpass both a probability threshold and a minimum confidence interval based on recent volatility. Instead of investing a fixed amount per trade, I would use a dynamic allocation based on the confidence score: allocate more capital to higher-confidence predictions using a weighted formula. Additionally, I would implement a stop-loss rule to automatically sell any position if the stock drops more than 5% after purchase, to limit downside risk. This strategy would reduce the number of trades, concentrate on high-confidence opportunities, and actively manage risk while aiming to maximize returns. 1
Use multiple models (e.g., logistic regression, random forest, XGBoost) and aggregate their predicted probabilities using an average or weighted ensemble. Select only the top 3 predictions with the highest average confidence scores above a minimum threshold of 0.75 — this filters out low-confidence trades. Allocate available capital proportionally to each trade based on its confidence score. For example, a prediction with 0.9 confidence gets a larger share than one with 0.76. Cap total trades per round to 3 to reduce overtrading and manage risk. Implement a stop-loss or performance review — if any prediction results in a loss greater than 20%, reduce its model weighting in future rounds. 1
Use a regra mais precisa da Random Forest (limite ≥ 0,63), limitada a 5–7 negociações/dia, para sinais de alta qualidade. 1
My ideal trading strategy would focus on high-confidence and risk-adjusted returns. I would filter predictions with a probability above 0.75 and select the top 5 companies with the highest Sharpe Ratios to prioritize both expected growth and stability. Funds would be allocated proportionally based on each company's Sharpe score, giving more weight to those with better risk-adjusted performance. To reduce overtrading and manage risk, I would limit myself to no more than 3 trades per week, rebalance the portfolio monthly, and ensure diversification across different sectors. 1
My ideal strategy would focus on selective, high-confidence trades using multiple thresholds and models. Specifically, I would: Combine predictions from the best-performing random forest and decision tree models. Filter for high-confidence signals by selecting only those predictions with a probability score above 0.70. Limit the number of trades per day by choosing only the top 2 predictions (by score) to avoid overtrading and reduce fees. Allocate capital dynamically, investing a larger share (e.g., 70%) in the top-ranked signal and 30% in the second-best, rather than equally. To reduce drawdowns, I would skip trades on days when the overall market (S&P 500) is down, using a market trend filter. This approach combines model strength, confidence-based filtering, risk control, and capital efficiency. I would implement it by merging predictions into a daily decision table, ranking them, applying thresholds and filters, and allocating capital accordingly. 1
1) Receive best fee/'one time investment' available from my broker. 2) Take model with average ~8 trades per day (or less if fees will be high). For manually checking all operations during day. And slowly increase number of trades to 24 for rarely checks in the future. 3) Invest everything proportionally strong future growth. Not exceeding number of trades. And not exceeding optimal fee. 4) 'Everything' means some quantity for this level of trust to model. 5) Mark model with level of trust score. 6) Slowly increase level of trust and increase overall investment value. 1
My Strategy: Daily Monitoring & Entry Criteria: Track a predefined list of stocks daily to identify optimal entry points for trades. Each position is intended to be held for up to 30 days, targeting a predefined profit level upon exit. Exit Strategy & Model Integration: For acquired stocks, an additional predictive model may be employed to assess potential downward price movements, helping determine the optimal selling point. Two scenarios will be simulated to evaluate risk management: Scenario 1: Selling stocks at a loss if they underperform beyond a defined threshold. Scenario 2: Holding underperforming stocks on the balance sheet, anticipating eventual recovery and profitability. Objective: Assess the effectiveness of both exit approaches to refine risk-adjusted returns and optimize the strategy. 1
Take the 3 best (above 0.07) split equally long, take the 3 worst (below 0.02) split equally short, assume this can be leveraged 2x (i.e. from an initial available cash of 100 - I am 200 long and 200 short) 1
I would use predictions from two different models to gain diverse perspectives on each trading opportunity. For each trading period, I would: Select assets where both models predict high probability (e.g., both have a prediction above 0.75), ensuring stronger consensus and reducing false positives. Of these, choose the top 5 assets with the highest average predicted probability. Allocate my available capital equally among these 5 trades for diversification. To minimize overtrading, only invest if at least three assets meet the criteria. Implement a trailing stop-loss on each position (e.g., 5%) to automatically limit any losses, and set a take-profit limit (e.g., 10%) to lock in gains. Re-evaluate predictions and portfolio weekly to adjust positions based on updated model outputs and market conditions. 1
My ideal trading strategy is to select the top 5 predictions with probabilities above 0.8 and invest equal amounts in each. If fewer than 5 meet the threshold, I invest only in those that do. I rebalance weekly and hold positions for one week. To reduce overtrading and manage risk, I include a cooldown period of one week after exiting a position, during which I do not reenter the same prediction. 1
My ideal trading strategy would focus on selecting only high-confidence predictions to maximize returns while limiting risk. I would invest in the top 5 predictions with probabilities above 0.8, distributing my total capital equally among them. To reduce unnecessary trades, I would only rebalance the portfolio weekly instead of daily. Additionally, I would set a stop-loss at 5% below the purchase price for each trade to minimize losses and protect gains. 1
My ideal strategy would focus on maximizing returns while managing risk through selective investing and dynamic allocation. I would combine predictions from two different models to increase confidence. I would filter trades by only selecting opportunities where both models agree and the average confidence score is above 0.8. From these, I would pick the top 5 predictions with the highest scores and invest proportionally based on their confidence—higher confidence gets a larger share of capital. To reduce the number of trades, I would limit investments to a maximum of 5 per cycle and implement a stop-loss of 10% to manage downside risk. This approach aims to balance selectivity, diversification, and risk control. 1
I would filter trades by only selecting opportunities where both models agree and the average confidence score is above 0.8. From these, I would pick the top 5 predictions with the highest scores and invest proportionally based on their confidence; higher confidence gets a larger share of capital. 1
My ideal trading strategy combines high-confidence predictive signals—such as threshold-based models with scores above 7 or 8—with a selection of momentum and trend indicators. I also incorporate metrics from the options market to gain insight into sentiment and volatility. I favor using Fibonacci retracements, as an example, to identify potential reversal zones, and rely on support and resistance levels to validate entries and define stop-losses and exits. This helps ensure trades are grounded in both statistical and structural market logic. My money management approach is adaptive, meaning position sizes change based on portfolio performance and perceived risk. I scale into strength and reduce exposure during drawdowns, aligning capital deployment with market conditions. Given limited capital, I track positions individually and use maximum drawdown as a key metric in backtesting. I follow a “less is more” philosophy—fewer trades, but higher quality setups—with active risk management as a core principle. Options play a central role in the portfolio. I use them to capture volatility, ride trends, or hedge stock positions. This adds flexibility and leverage while enhancing risk control. I rely on quantitative tools and diverse data feeds to support both stock and options analysis. This includes machine learning models—especially neural networks—to improve predictive accuracy and uncover nonlinear relationships in market data. I plan to expand their use further as part of my strategy development. The goal is not to trade more, but to trade smarter—focusing on consistent, risk-adjusted returns with strong capital preservation and precision execution. 1
Description of the Ideal Trading Strategy. My ideal strategy is an "Adaptive Portfolio with Multi-Layered Risk Management": 1. Prediction Selection: I combine predictions from both the Random Forest and Decision Tree models, only acting on signals where both models are in agreement and the probability exceeds 0.65. Additionally, I apply a filter based on macroeconomic conditions (DGS10, FEDFUNDS) to confirm the market environment. 2. Dynamic Capital Allocation: Instead of a fixed $100, I invest an amount proportional to the model's confidence: 70-75% probability: 1x the base amount 75-85% probability: 1.5x the base amount 85%+ probability: 2x the base amount 3. Limiting the Number of Trades: I select a maximum of the 5 best predictions per day to reduce transaction costs and improve focus on high-quality signals. 4. Risk Management: I implement a stop-loss at -12% for a quick exit from losing positions, regardless of the 30-day ML prediction horizon. I also use position sizing, allocating no more than 10% of my capital to a single position. 5. Reinvestment: Profits from closed positions are reinvested to compound returns, while losses reduce the available capital for new trades. This strategy combines the best aspects of the course: high-quality ML predictions, risk management, efficient use of capital, and the practical constraints of real trading. 1

Calculated: 5 August 2025, 22:04