Home/Concepts/Finance/Algorithmic Trading Basics

Algorithmic Trading Basics

Build and backtest simple trading strategies

⏱️ 24 min⚡ 9 interactions

1. Trading at the Speed of Light

In 2010, a single algorithmic trading error caused the "Flash Crash," wiping out $1 trillion in market value in minutes. Today, over 70% of US equity trades are executed by algorithms. These bots make split-second decisions using mathematical models, processing millions of data points per second.

🤖 Core Concept

Algorithmic trading uses computer programs to execute trades based on predefined rules. The three main types are: Momentum strategies (trend following), Mean reversion (betting on price corrections), and Arbitrage (exploiting price differences). Success requires backtesting, low latency, and rigorous risk management.

🎯 Interactive: Choose Trading Strategy

Sharpe Ratio
1.2
Max Drawdown
15%
Win Rate
55%
Trades/Year
45

2. Building a Moving Average Crossover Strategy

🎓 From Idea to Algorithm: Building Trading Logic

Moving Average Crossovers: The Hello World of Algo Trading

Moving average (MA) crossover strategies are the simplest trend-following systems. The logic: when a fast MA (e.g., 10-day) crosses ABOVE a slow MA (e.g., 50-day), BUY (trend turning up). When fast crosses BELOW slow, SELL (trend turning down).

MA Crossover Logic:
Golden Cross (BUY Signal) 📈
Condition:
IF (yesterday: Fast MA <= Slow MA) AND (today: Fast MA > Slow MA)
→ Enter LONG position
Example: 10-day MA crosses above 50-day MA on SPY at $450
Interpretation: Short-term momentum stronger than long-term trend
Action: Buy 100 shares, set stop-loss at 50-day MA ($445)
Historical win rate: ~55% in trending markets, ~40% in choppy markets
Death Cross (SELL Signal) 📉
Condition:
IF (yesterday: Fast MA >= Slow MA) AND (today: Fast MA < Slow MA)
→ Exit LONG position (or enter SHORT)
Example: 10-day MA crosses below 50-day MA on SPY at $455
Interpretation: Short-term momentum weakening, trend reversing
Action: Sell position, capture $5 profit ($455 - $450 entry)
Death crosses preceded 70% of major market corrections since 2000
Why this works (when it works): MAs smooth out price noise, revealing the underlying trend. Crossovers identify momentum shifts early. BUT—they lag price (by design), so you miss the first 10-20% of moves and get whipsawed in sideways markets. It's NOT a holy grail, just a simple starting point.

Parameter Optimization: The Overfitting Trap

Should you use a 10/50 MA crossover? 5/20? 20/200? Optimization finds the best historical parameters—but also creates overfitting: a strategy that perfectly fits past data but fails on new data. This is the #1 killer of algo strategies.

Optimization Results: SPY (2015-2020)
Fast/SlowTotal ReturnWin RateMax DDTrades
5 / 20+45%52%-18%38
10 / 50+38%58%-12%22
20 / 100+28%62%-9%12
⭐ 13 / 37+62%68%-7%28
50 / 200+18%55%-11%5
⚠️ The Overfitting Problem:
The 13/37 combo looks PERFECT: +62% return, 68% win rate, low drawdown! You're rich, right?
WRONG. Test it on 2020-2025 data:
13/37: -12% return (worst performer!)
10/50: +22% return (best performer)
Why? The 13/37 parameters fit random noise in 2015-2020 data, not real market structure. It was "curve-fitted" to past quirks that didn't repeat.
✅ How to Avoid Overfitting:
• Walk-forward analysis: Optimize on Year 1-3, test on Year 4, re-optimize, test Year 5, repeat
• Out-of-sample testing: Keep 30% of data hidden until final test (never use it for optimization)
• Simple parameters: Round numbers (10, 20, 50, 100, 200) are more robust than weird ones (13, 37, 83)
• Reasonable performance: If backtest shows 80% win rate or 100%+ annual return, it's probably overfit!
Rule of thumb: Live performance = 50-70% of backtest performance (after costs, slippage, reality)
Professional approach: Use robust parameters (10/50, 20/100) that have been widely used for decades. If thousands of traders use them, they're harder to arbitrage away. Fancy optimized parameters might give you a slight edge for 6 months before decaying. Simplicity + consistency > complexity.

Backtesting: The Reality Check

Backtesting runs your strategy on historical data to see what WOULD have happened. But backtests lie! They don't include real-world friction: commissions, slippage, liquidity constraints, and execution delays.

Backtest vs Reality: What Gets Lost
Perfect Backtest (Unrealistic):
Strategy return:+50%/year
Win rate:65%
Sharpe ratio:2.1
Trades executed:100% at desired price
↓ Reality Tax ↓
Live Trading Reality:
Commissions (120 trades × $1):-$120 (-0.12%)
Slippage (0.1% per trade):-12% total
Missed fills (10% of trades):-5% opportunity
Latency (50ms delay):-3% timing
ACTUAL return:+29.8%/year
ACTUAL Sharpe:1.2
From +50% backtest to +30% reality = 40% performance decay!
📊 Realistic Backtest Checklist:
✅ Include commissions ($0.50-$5 per trade depending on broker)
✅ Add slippage (0.05-0.2% per trade for liquid stocks, 0.5%+ for illiquid)
✅ Model bid-ask spread (enter at ask, exit at bid = instant 0.1% loss)
✅ Add latency delay (50-200ms for retail, 5ms for institutions)
✅ Test across multiple market regimes (bull, bear, sideways, volatile)
✅ Ensure enough liquidity (can't buy $1M of microcap at market price!)
If strategy survives all this and still makes 15-20%+ annually, you might have something real!
The 50% rule: Assume your live trading returns will be ~50% of backtest returns. Backtest shows +40%? Expect +20% live (still excellent!). Backtest shows +10%? Expect +5% live (barely worth the effort after taxes). This is why HFT needs 100%+ backtest returns—live they make 20-30% after costs, which is still great on billions of capital.

📊 Interactive: Strategy Parameters

10 days
30 days
Price
Fast MA (10)
Slow MA (30)
Total Trades
1
Win Rate
0.0%
Avg Return
0.00%
Signals
1

🔄 Interactive: Backtesting Period

252 days (1.0 years)
💡 Longer backtests = more reliable: Test across multiple market cycles. A strategy that works in bull markets might fail in bear markets. Minimum 3-5 years recommended.

3. Order Execution & Slippage

🎓 Market Microstructure: Where Profits Go to Die

The Three Order Types (And Why Each One Costs You)

Every algo strategy eventually asks: "How do I actually BUY this stock?" The answer determines whether your 20% backtest becomes 15% reality or -5% disaster. Order type is the tradeoff between speed (certainty of execution) and price (slippage cost).

Order Type Comparison:
1. Market Order (Speed > Price)
What it is:
"Buy NOW at whatever price is available"
How it executes:
Hits the best ask price on the order book instantly (or best bid for sells)
Pros:
✅ Guaranteed execution
✅ Instant (milliseconds)
✅ Simple logic
Cons:
❌ Pay bid-ask spread (0.1%+)
❌ Slippage on large orders
❌ Worst price possible
Example: Stock bid $100.00, ask $100.10. Market buy order executes at $100.10 (you pay 0.1% spread). On $100K order, that's $100 lost instantly!
2. Limit Order (Price > Speed)
What it is:
"Buy ONLY if price reaches $X or better"
How it executes:
Sits on order book at your limit price, waits for seller to match you
Pros:
✅ Control price (no slippage)
✅ Can capture spread
✅ Maker rebates (some venues)
Cons:
❌ May never fill (miss trade!)
❌ Partial fills possible
❌ Slower (seconds to hours)
Example: Stock at $100.10, you place limit buy $100.00. If price drops to $99.95, you buy at $100.00 (saved $0.10). But if price rallies to $102, you miss the trade entirely!
3. Stop Order (Trigger-Based)
What it is:
"If price hits $X, submit market/limit order"
How it executes:
Dormant until trigger price, then becomes market order (stop-loss) or limit order (stop-limit)
Pros:
✅ Automated risk management
✅ No monitoring needed
✅ Catches breakouts
Cons:
❌ Slippage when triggered
❌ Can be "hunted" by HFTs
❌ Gap risk (trigger at $95, fill at $90!)
Example: Buy at $100, set stop-loss at $95. Stock gaps down to $92 overnight. Your stop triggers at market open but fills at $92 (-8% loss, not -5%!).
Pro strategy: Use limit orders for entries (wait for good price), market orders for urgent exits (get out NOW in crashes). HFT uses 99% limit orders (they provide liquidity, earn rebates). Retail uses 70% market orders (pay for convenience). That's one reason HFT is profitable!

Slippage: The Silent Killer

Slippage is the difference between expected price and execution price. It's invisible in backtests (assumes perfect fills) but brutal in live trading. On a 100-trade strategy, 0.1% average slippage = -10% annual return gone!

Sources of Slippage:
1. Bid-Ask Spread (Always There)
SPY (mega-liquid):0.01% spread
AAPL (large-cap):0.05% spread
Mid-cap stock:0.15% spread
Small-cap / illiquid:0.5-2% spread
Market orders pay the full spread. On 100 trades/month in mid-caps: 0.15% × 100 = 15% annual cost!
2. Market Impact (Large Orders)
Scenario:
Order book: 100 shares @ $100.00, 200 @ $100.05, 500 @ $100.10
Your order: Buy 1,000 shares market
Execution:
• 100 shares @ $100.00
• 200 shares @ $100.05
• 500 shares @ $100.10
• 200 shares @ $100.15 (had to go deeper!)
Average fill: $100.096 (0.096% slippage from $100.00)
Rule: If your order is >1% of average daily volume, expect significant market impact (0.5-5% slippage)
3. Latency (Network + Processing Delay)
You see price: $100.00 → Click BUY → 50ms network → 20ms broker → 30ms exchange
Total 100ms delay. In that time, HFT moved price to $100.08!
HFT (1-5ms latency):0.01% slippage
Institutions (10-50ms):0.03% slippage
Retail (50-200ms):0.1% slippage
Slow internet (500ms+):0.5%+ slippage
This is why HFT firms co-locate servers at exchanges. 1ms advantage × 1,000 trades/day = major edge.
📊 Total Slippage Calculator:
Example: Day trading mid-cap stocks (100 trades/month)
Bid-ask spread: 0.15% per trade
Market impact: 0.05% (small positions)
Latency slippage: 0.08% (retail connection)
= 0.28% per round-trip trade
× 100 trades = 28% annual cost!
Your strategy needs 30%+ backtest returns just to break even after slippage! This is why 90% of active traders lose money.
Minimizing slippage: Trade liquid stocks (SPY, AAPL, MSFT have 0.01% spreads), use limit orders (capture spread instead of paying it), reduce trade frequency (100 trades/year not 100/month), use algo execution (TWAP/VWAP slice large orders), and NEVER market order on open/close (spreads widen 5-10×!).

📋 Interactive: Order Types

0.1%
Requested Price:$100.00
Executed Price:$100.02
Slippage:$0.023 (0.02%)
Execution Latency:50ms

⚙️ Interactive: Execution Algorithms

TWAP Strategy

Splits a large order into equal slices executed at regular intervals. Example: 10,000 shares over 10 minutes = 1,000 shares per minute.

⚡ Interactive: Latency & HFT

50ms
1ms (HFT)50ms (Retail)200ms (Slow)
1-5ms
HFT
Co-located servers
5-20ms
Low-Latency
Fiber optic
20-100ms
Retail
Standard brokers
100+ms
Slow
Disadvantaged
⚡ Speed advantage: At 50ms latency, you're fast enough for day trading. HFT firms pay millions for 1-2ms speed improvements through co-location and microwave networks.

4. Risk Management & Trading Frequency

🎓 The Frequency Paradox: More Trades ≠ More Profit

Trading Frequency Tiers: The Reality Behind the Speed

Algo trading spans from low-frequency (hold days/weeks, 10 trades/month) to high-frequency (hold seconds, 10,000 trades/day). Higher frequency requires exponentially more capital, infrastructure, and sophistication. Most successful retail traders stay low-frequency.

Frequency Tier Comparison:
Low Frequency (Swing Trading) 📊
Characteristics:
• Hold time: 1-30 days
• Trades: 5-50/month
• Latency tolerance: 100ms+
• Profit/trade: 1-5%
Requirements:
• Capital: $10K-$100K
• Infrastructure: Retail broker
• Data: EOD or 1-min bars
• Cost: $100/month
✅ Advantages:
• Low transaction costs (0.5-1% annual)
• Manageable with day job
• Less infrastructure needed
• Captures real trends (not noise)
❌ Disadvantages:
• Overnight risk (gaps)
• Lower Sharpe ratio (0.5-1.5)
• Emotional discipline needed
Example: 20-trade/month MA crossover strategy. $50K account, 1% risk/trade = $500 risk. With 55% win rate and 2:1 R:R, expect 15-25% annual return after costs.
Medium Frequency (Day Trading) ⚡
Characteristics:
• Hold time: Minutes to hours
• Trades: 20-200/day
• Latency tolerance: 20-100ms
• Profit/trade: 0.1-0.5%
Requirements:
• Capital: $50K-$500K (PDT rule!)
• Infrastructure: DMA broker
• Data: Tick data, L2 quotes
• Cost: $1K-$5K/month
✅ Advantages:
• No overnight risk (flat by close)
• More opportunities (50-100 trades/day)
• Higher Sharpe potential (1.5-2.5)
❌ Disadvantages:
• Transaction costs 5-10% annual
• Full-time commitment required
• PDT rule ($25K minimum)
• High stress, burnout risk
Example: 50-trade/day scalping strategy. $100K account, need <$2 commission/trade and 0.05% slippage to be profitable. Win rate 52%, avg 0.15% per winner. Gross 30%, net 20% after costs.
High Frequency (HFT) 🚀
Characteristics:
• Hold time: Milliseconds to seconds
• Trades: 1K-1M+/day
• Latency tolerance: 1-10ms
• Profit/trade: 0.001-0.01%
Requirements:
• Capital: $1M-$100M+
• Infrastructure: Co-location
• Data: Direct exchange feed
• Cost: $50K-$500K/month
✅ Advantages:
• Massive scale (billions traded/day)
• High Sharpe (2.5-4.0+)
• Market-neutral (no overnight risk)
• Statistical edge vs slower traders
❌ Disadvantages:
• Requires PhDs + massive capital
• Arms race (speed constantly decaying)
• Regulatory scrutiny
• Alpha decay 30%+/year
Example: Market-making on SPY. 100K trades/day, $0.01 profit/trade = $1,000/day gross. But costs: $100K/month infrastructure + $50K/month salaries. Need $5B volume/day to be profitable!
Reality check: 95% of retail traders should stick to low-frequency. Medium requires $50K+ and full-time focus. HFT is unreachable (you're competing with Citadel, Virtu, Jane Street with billions in capital). The "trade more = earn more" myth kills accounts—higher frequency = higher costs that eat all profits unless you have infrastructure + edge.

Cost Structure: Why Frequency Matters

Every trade has fixed costs (commissions) and variable costs (slippage). Higher frequency multiplies these costs exponentially. A 0.1% cost/trade strategy making 10 trades/month = -1.2% annual cost (manageable). Same strategy making 1,000 trades/month = -120% annual cost (impossible!).

Annual Cost Analysis by Frequency:
Cost ComponentLow (10 trades/mo)Med (100 trades/mo)High (10K trades/mo)
Commissions-0.1%-0.6%-5%
Slippage-0.5%-5%-30%
Data Feeds-0.1%-1%-5%
Infrastructure-0.1%-2%-10%
Opportunity Cost-0.2%-3%-15%
TOTAL COST-1.0%-11.6%-65%
⚠️ The High-Frequency Trap:
HFT strategy needs 65%+ gross returns just to BREAK EVEN after costs! Here's why most fail:
• Slippage scales with volume: 10K trades/month = 20K trade sides = massive market impact
• Infrastructure costs: Co-location $10K/month, data feeds $20K/month, servers $15K/month
• Opportunity cost: Could've earned 10% risk-free in bonds instead of stressing over microseconds
Unless you're Citadel with billions and PhDs, stick to low-frequency!
The sweet spot: Low-frequency strategies (10-50 trades/month) have manageable 1-2% annual costs, leaving room for 15-25% net returns. Medium-frequency (100 trades/month) needs exceptional edge to overcome 10%+ costs. High-frequency needs institutional resources. Don't fight the math—trade less, earn more!

Position Sizing: Risk of Ruin in Algo Trading

Algos don't get emotional—but they can still blow up accounts with bad position sizing. The risk of ruin formula shows how many consecutive losses wipe you out. With 5% risk/trade, you're broke in 14 losses. With 1%, you can survive 70 losses!

Risk of Ruin Analysis:
Formula: Losses to 50% Drawdown
N = log(0.5) / log(1 - Risk%)
Where N = consecutive losses needed to lose 50% of account
0.5% risk/trade:138 losses to ruin
1% risk/trade:69 losses to ruin
2% risk/trade:35 losses to ruin
5% risk/trade:14 losses to ruin
10% risk/trade:7 losses to ruin 💀
⚠️ Probability of Consecutive Losses:
Even with 60% win rate strategy (pretty good!), probability of losing streaks:
5 losses in a row:1.0% chance (happens!)
7 losses in a row:0.16% chance (rare but possible)
10 losses in a row:0.01% chance (unlikely but not impossible)
With 10% risk/trade, just 7 losses = account blown. That 0.16% chance WILL happen eventually!
✅ Professional Position Sizing Rules:
• Retail traders (small accounts): Use 0.5-1% risk per trade max
• Full-time algo traders: Use 1-2% risk per trade max
• Institutions (diversified): Can use 2-3% risk per strategy (but run 10+ uncorrelated strategies)
Golden rule: Size so that 20 consecutive losses = 20% drawdown (recoverable). For 60% win rate, that's ~1% risk/trade.
Kelly Criterion adjustment: Pure Kelly often suggests 5-10% position sizes (way too aggressive!). Use Quarter Kelly or Half Kelly for algo trading. If Kelly says 8%, use 2% (quarter) or 4% (half). This massively reduces drawdowns while keeping 70-90% of Kelly growth. Survival > optimization!

🔄 Interactive: Trading Frequency

Commission Impact
2%
Infrastructure Cost
$1K/mo
Avg Holding Time
Hours
Complexity
Medium

🎲 Interactive: Position Sizing

1%
Consecutive Losses to Ruin (50% drawdown)
69
losing trades in a row
Ultra Safe (0.5%)
139 losses = 50% drawdown
139
Conservative (1%)
69 losses = 50% drawdown
69
Moderate (2%)
35 losses = 50% drawdown
35
Aggressive (5%)
14 losses = 50% drawdown
14

📊 Interactive: Strategy Performance

StrategyWin RateSharpeMax DDComplexity
Momentum55%1.215%Low
Mean Reversion60%0.912%Medium
Arbitrage85%2.15%High
Market Making52%1.88%Extreme
Statistical Arb58%1.510%High

5. Key Takeaways

🔄

Backtest Everything

Never trade a strategy without backtesting it on at least 3-5 years of data across multiple market conditions. A strategy that works in bull markets might crater in bear markets. Out-of-sample testing is crucial.

💸

Transaction Costs Kill Returns

Commissions, slippage, and market impact can turn a profitable backtest into a losing live strategy. High-frequency strategies need commissions <$0.001/share. Include realistic costs in backtests.

⚡

Speed = Edge in HFT

High-frequency trading requires sub-5ms latency. Firms pay millions for co-location at exchanges and microwave networks between cities. Retail traders can't compete at this speed—focus on slower strategies.

📊

Overfitting Destroys Strategies

Optimizing parameters on historical data can create a strategy that perfectly fits the past but fails in live trading. Use walk-forward analysis and keep strategies simple with few parameters.

🛡️

Risk Management is Non-Negotiable

Even with 60% win rate, risking 5% per trade = ruin in 14 losses. Use 0.5-2% risk per trade maximum. Implement circuit breakers to stop trading after X losses. One bad day shouldn't wipe out your account.

🎯

Alpha Decays Quickly

Profitable strategies attract competition until returns decline. The average algo strategy's edge decays 20-30% per year as others discover it. Continuously research and adapt or returns will vanish.