7. Execution Algorithms

Once your size grows past a certain threshold, how you enter and exit matters as much as what you trade. A good signal poorly executed loses money. This chapter is about minimizing market impact and slippage.

The execution problem

You want to buy 50,000 shares of XYZ.

• Average daily volume = 200,000 shares.
• Best ask currently = ₹500, ask depth = 2,000 shares.

If you blast a market order:

• First 2k fill at 500.
• Next 3k at 500.50.
• Next 5k at 501.20.
• …You finish around 504, paying 0.8% in slippage (~₹2L).

Smart execution can reduce that to 0.1–0.2%.

Order types — execution focused

Order	Behavior	Use
Market	Eats book immediately	Tiny orders, urgency
Limit (passive)	Sits on the book	Patient entries, earns spread
Limit (aggressive)	Crosses spread	Takes liquidity, controlled fill
IOC (Immediate or Cancel)	Fills what’s available now, cancels rest	Multi-leg, hidden execution
FOK (Fill or Kill)	All-or-nothing instant	Rare, usually for blocks
Iceberg	Shows only a slice; refills automatically	Hide size from the book
AMO	Queues for next session	Off-hours decisions

In NSE retail, iceberg is supported by most brokers (with ~10 disclosed slices max).

TWAP — Time-Weighted Average Price

Spread the order evenly over time.

Total qty: 50,000 over 60 minutes
Slice every minute: ~833 shares

• Pros: Predictable, low signaling.
• Cons: Doesn’t adapt to market conditions; bad in trending markets.
• Use: Simple, when you don’t have a strong view on intraday liquidity.

VWAP — Volume-Weighted Average Price

Match the historical intraday volume curve.

NSE’s typical volume profile (U-shaped):

• 09:15–10:00: 22% of day’s volume
• 10:00–14:00: 30% (sparse)
• 14:00–15:30: 48% (afternoon surge)

A VWAP algo executes more shares when volume is high → blends in with natural flow.

• Pros: Better than TWAP in normal markets, common benchmark.
• Cons: Predictable if others know your style; vulnerable to gaming.
• Use: Standard for institutional execution. Brokers offer VWAP algos.

POV — Percentage of Volume

Trade a constant % of the market’s real-time volume.

target_pov = 10%
if last_minute_market_volume == 5000:
    my_order = 500 in next minute

• Pros: Adapts to market; never dominates.
• Cons: May not finish if volume is low; can take all day.
• Use: Large orders where finishing on time is flexible.

Implementation Shortfall (IS)

Minimize total cost vs the arrival price (price when decision was made).

Cost = Market impact + Timing risk + Spread cost.

IS algos balance speed (less timing risk) vs patience (less impact). A more aggressive IS algo trades fast (paying impact). A more passive one trades slow (eating timing risk).

Used by quants who care about beating arrival price, not VWAP.

Iceberg orders (deeper)

You want to buy 10,000 but show only 500 at a time.

• Place limit at ₹500 with disclosed qty 500.
• When 500 fills, broker auto-places another 500.
• Repeat until total filled.

Avoids signaling your size to other participants. Especially useful in mid/small caps with thin books.

Indian exchanges have minimum disclosed qty rules — check broker docs.

Sniper / opportunistic execution

For mean-reversion entries: place a limit deep in the book and wait. If price reaches you, you got a great fill. If not, you missed — but no slippage.

Stock at 500.10
Bid stack: 100@500.00, 200@499.95
You place: buy 1000 @ 499.50

Wait... if dipped, you got filled at 499.50.
If not, no trade.

Used by HFTs and passive market makers. Adds liquidity, earns spread.

Slippage modeling

For backtesting realism, slippage cost ≈ a function of:

$\text{slippage} \approx k \cdot \sigma \cdot \sqrt{\frac{Q}{V}}$

Where $\sigma$ = volatility, $Q$ = your order size, $V$ = ADV. Empirically, $k \approx 0.5–1.0$.

For most retail orders < 0.1% of ADV → slippage is just spread + 0.05%. Above 1% of ADV → it becomes painful.

Market impact — temporary vs permanent

• Temporary impact — price moves due to your order pressure, then reverts after you stop.
• Permanent impact — your trade revealed information; price stays moved.

Larger orders = larger permanent impact. Splitting orders over time reduces both, but doesn’t eliminate them.

Choosing an execution style

Order size vs ADV	Recommended
< 0.1%	Market or aggressive limit, single shot
0.1% – 1%	Iceberg or simple TWAP over 5–30 min
1% – 5%	VWAP or POV over the day
> 5%	Multi-day execution; consider negotiating block trade

Most retail traders never need more than market/limit + iceberg. Algos matter when you’re executing crores per trade.

Execution metrics to track

For every fill:

• Slippage vs decision price (basis points).
• Slippage vs VWAP.
• Fill rate (filled / requested).
• Reversion — did price come back after your trade? (Indicates over-aggressive execution.)

Average over hundreds of trades → know your real cost of trading, not the theoretical zero of backtests.

Direct Market Access (DMA) & broker APIs

For algo execution, you need:

• Low-latency broker API (Kite Connect, Upstox, Angel One — all decent for retail).
• Co-location for HFT (only worthwhile for sub-millisecond strategies; institutional only).
• Reliable internet + UPS + 4G failover.

Most retail trading bots (including this project) run on standard broker APIs, polling at 1–5 second intervals — completely fine for swing/positional strategies.

Common execution pitfalls

1. Market orders in illiquid stocks — get fleeced. Always use limits.
2. Blasting size at the open — first 5 minutes have wide spreads & noise. Wait.
3. Chasing fills — modifying limit upward after misses → ends with worst price of the day.
4. Ignoring corporate actions — orders may behave oddly on ex-date.
5. Stale prices — using delayed quotes (15-min lag) for execution decisions.
6. No fill timeouts — orders left dangling can fill at terrible prices later.

Testing execution

Before going live with a new algo:

1. Paper trade with simulated execution that includes realistic slippage.
2. Live test with 1 share — verify the API does what you expect.
3. Live test with small size — measure actual fills vs expectations.
4. Scale up gradually — only after 100+ trades of consistent behavior.

Bugs in execution can lose money faster than bugs anywhere else. Have hard sanity checks: max order size, max # orders per minute, kill switch on unexpected behavior.

Reading list

• Algorithmic Trading & DMA — Barry Johnson (the standard reference).
• Trades, Quotes and Prices — Bouchaud, Bonart, Donier, Gould.