Lesson 02: Financial Markets and Trading Basics

If your Agent doesn't understand order books, slippage, and market impact, it's just a machine that writes essays.

An Expensive Lesson

A quant newbie developed a strategy that looked very profitable: backtested annual return of 80%, Sharpe ratio of 2.0. He excitedly went live.

In the first week, the strategy signal said "buy 1000 shares of Tesla," and he did. The backtested expected cost was $250,000 ($250 x 1000 shares).

Actual execution price? $253,500.

He lost $3,500 before even starting to make money.

What happened?

1. Slippage: His market order "ate through" multiple price levels in the order book, with an average price $2.5/share higher than expected
2. Fees: Commission + SEC fees = approximately $50 loss per trade
3. Market Impact: His large order attracted attention from other traders, pushing up the price

Three months later, the strategy's live performance was 40% lower than the backtest. The strategy wasn't bad - he just miscalculated the costs.

Goal of this lesson: Help your Agent understand how real markets work, and avoid the tragedy of "profitable in backtest, losing money live."

2.1 Market Types

Characteristics of Different Markets

Implications for Agents:

• Crypto 24/7 -> Agent must run around the clock, requiring stronger automation
• Futures high leverage -> Risk Agent's stop-loss logic must be stricter
• China A-shares T+1 (regular stocks bought today can only be sold tomorrow, some ETFs excepted) -> Execution Agent needs to consider overnight risk

Primary Market vs Secondary Market

This course focuses on the secondary market - the main battlefield for quantitative trading.

Exchanges and Matching Mechanisms

Whether it's US stocks, China A-shares, or cryptocurrencies, all trades ultimately flow to the Matching Engine.

The exchange isn't your counterparty; it's just a "matchmaker," matching buyers and sellers through rules:

Matching Rules (most exchanges):
1. Price Priority: Higher bid orders get filled first
2. Time Priority: At the same price, earlier orders get filled first

Multi-Agent Perspective:

• Execution Agent: Must understand matching logic, deciding between market orders and limit orders
• Research Agent: Identifies large fund flows through transaction data

2.2 Basic Trading Units

Asset

An asset is what you trade. Different assets have different code conventions:

Time Scales: Tick -> Candlestick

Quant systems process data at different granularities:

Tick Data (finest)
    | aggregate
Minute Bars (1min, 5min, 15min)
    | aggregate
Daily / Weekly / Monthly (coarsest)

Tick Data: Every trade, every quote change. Essential for high-frequency strategies, high storage cost.

OHLCV (Candlesticks): Standard format after aggregating ticks:

• Open (opening price)
• High (highest price)
• Low (lowest price)
• Close (closing price)
• Volume (trading volume)

Order Book

The order book shows market "depth" - how many orders are queued at different price levels:

Spread (Bid-Ask Spread) = Best Ask - Best Bid = $185.01 - $184.99 = $0.02

• Smaller spread -> Better liquidity -> Lower trading costs
• Larger spread -> Worse liquidity -> Higher slippage risk

2.3 The Real Impact of Trading Costs

Backtested 50% annual return, losing money live? Trading costs are the culprit.

Slippage

Backtests assume you can buy at the best price, but in live trading:

You want to buy 1000 shares of AAPL, order book:
  $185.01 - 200 shares  <- You eat these 200 first
  $185.05 - 500 shares  <- Then eat these 500
  $185.10 - 300 shares  <- Finally eat these 300

Actual average price = (185.01x200 + 185.05x500 + 185.10x300) / 1000 = $185.057
Expected price = $185.01
Slippage = $0.047/share = Total $47

Fees: Cumulative Effect

Seemingly small fees get amplified in high-frequency trading:

# Assuming 0.1% fee per trade
fee_rate = 0.001
trades_per_day = 50
trading_days = 250

# Annualized fee cost
annual_fee = fee_rate * 2 * trades_per_day * trading_days  # buy and sell each
print(f"Annualized fee cost: {annual_fee:.1%}")  # Output: 25.0%

25% annualized fee cost - this means your strategy's annual return must exceed 25% just to break even!

Market Impact

Your large order itself changes the market. Square Root Law estimation:

Market Impact ~ Y x sigma x sqrt(Q/V)

Y = constant (typically 0.5-1.0)
sigma = daily volatility
Q = your order size
V = average daily volume

def estimate_market_impact(order_size, daily_volume, daily_volatility, Y=0.5):
    """Estimate market impact cost"""
    participation = order_size / daily_volume
    impact = Y * daily_volatility * (participation ** 0.5)
    return impact

# Example: Order size is 1% of daily volume
impact = estimate_market_impact(
    order_size=1_000_000,
    daily_volume=100_000_000,
    daily_volatility=0.02
)
print(f"Estimated market impact: {impact:.2%}")  # Output: 0.10%

Cost Summary

Multi-Agent Perspective: The Execution Agent's core responsibility is minimizing these three costs.

Paper Exercise: Is Your Strategy Really Profitable?

Scenario: You developed a US stock intraday strategy with the following backtest parameters:

Question: What will the live return be?

Step-by-Step Calculation:

Step 1: Calculate per-trade costs

Per-trade slippage cost = $50,000 x 0.03% = $____
Per-trade SEC fee = $50,000 x 0.00278% = $____ (sells only)

Step 2: Calculate daily costs

Daily slippage = $____ x 20 trades = $____
Daily SEC = $____ x 10 trades (sells) = $____
Daily total cost = $____

Step 3: Calculate annualized costs

Annual cost = $____ x 252 days = $____
Total trading volume = $50,000 x 20 x 252 = $252,000,000
Annual cost rate = $____ / $252,000,000 = ____%

Step 4: Calculate live return

Live annual return = 35% - ____% = ____%

Answer (calculate first, then check):

Reflection Questions:

1. If slippage drops to 0.01%, can the strategy still be profitable?
2. If trading frequency drops to 5 times per day, what happens?
3. What insights does this give you for strategy design?

2.4 Strategy Lifecycle

A complete trade flows through the multi-agent system from inception to completion:

Detailed Flow:

1. Signal Generation (Signal Agent)

   • "AAPL's MACD shows bullish divergence, recommend going long"

2. Risk Review (Risk Agent)

   • "Current total position is 60%, max single position is 10%, this trade can only be 10%"
   • May reject, reduce size, or approve

3. Order Execution (Execution Agent)

   • "Order too large, split into 10 smaller orders, send one every 30 seconds using TWAP algorithm"

4. Execution Monitoring (Monitor Agent)

   • "5th child order slippage exceeded threshold, pausing subsequent execution"
   • Real-time execution quality feedback

5. Position Management and Exit (Position Agent)

   • "Position up 5%, triggering trailing stop"
   • "Position down 2%, triggering stop-loss exit"
   • Loop complete

Each stage can be an independent specialized Agent - this is the advantage of multi-agent architecture: specialized division of labor, clear responsibilities, easier debugging.

Lesson Deliverables

After completing this lesson, you will have:

1. Understanding of Market Structure - Know the characteristics and constraints of different markets (stocks/futures/crypto)
2. Trading Cost Awareness - Can estimate the impact of slippage, fees, and market impact on strategies
3. Strategy Lifecycle Perspective - Understand the complete loop from signal to exit

Verification Checklist

Use these checkpoints to confirm you truly understand this lesson:

If you can do these:

• Cost calculation accurate -> You have cost awareness
• Draw order book execution process -> You understand market microstructure
• Draw complete lifecycle -> You have systems thinking

If you cannot:

• Re-read relevant sections
• Walk through examples with specific numbers (e.g., AAPL $185)
• Find more detailed explanations in extended reading

Key Takeaways

• Understand characteristics of different markets (stocks/futures/forex/crypto) and their strategy implications
• Master OHLCV and order book fundamentals
• Recognize the three trading cost killers: slippage, fees, market impact
• Understand the complete strategy lifecycle loop: Signal -> Risk -> Execution -> Monitoring -> Exit

Extended Reading

• Background: Exchanges and Order Book Mechanics - Deep dive into L1/L2/L3 data differences
• Background: HF Market Microstructure - Essential reading if you want to compete with HFT
• Background: Cryptocurrency Trading Characteristics - Unique challenges of 24/7 markets

Next Lesson Preview

Lesson 03: Math and Statistics Fundamentals

Markets move, but how do we quantify these movements? Why do we use "returns" instead of "prices"? What is a "fat-tailed distribution," and why can the normal distribution assumption blow up your account? Find out next lesson.