Algorithmic and high-frequency trading

Algorithmic trading dominates modern markets. ~80% of US equity volume is algorithmic. The breakdown: about half is execution algos (helping institutions trade with minimal market impact), about half is proprietary algos (taking positions for profit). The two have different goals and different mechanics.

Execution algorithms

Execution algos solve a specific problem: institutions want to buy or sell large quantities without moving the market against themselves. The institutional trader's job is not to predict price direction; it is to fill a parent order produced by the portfolio manager at a cost as close to a defined benchmark as possible. Each algorithm targets a different benchmark and has different trade-offs between market impact, timing risk, and information leakage.

VWAP — the most-used execution benchmark

                  Σ ( price_i  ×  volume_i )
  VWAP   =   ────────────────────────────────────
                       Σ volume_i

where:
  i           =  index running over every trade in the measurement window
                  (typically the full trading day)
  price_i     =  price at which trade i printed
  volume_i    =  shares (or contracts) executed at price_i
  Σ           =  summation across all trades i in the window
  Numerator   =  total dollar value traded in the window
  Denominator =  total volume (share count) traded in the window

Reading: VWAP is the average market price weighted by how much volume
traded at each level. A trader whose average fill price beats the
day's VWAP has executed 'better than average' from a benchmark sense.

A VWAP execution algorithm aims to fill a parent order at a price close to the day's VWAP by slicing the order into child orders sized proportional to the forecast intraday volume curve. Historical volume curves are bell-shaped with humps near the open and close; VWAP algos place more child orders during high-volume periods to track the benchmark.

TWAP — simpler, slower, lower information leakage

  TWAP    =   ( 1 / N )  ×  Σ price_i

where:
  N        =  the number of equal time intervals into which the
               execution window is divided (e.g. 78 five-minute slots
               in a 6.5-hour US trading day)
  price_i  =  the average market price during interval i
  Σ        =  sum across the N intervals

A TWAP execution slices the parent order into equal child orders
delivered at equal time intervals. Each child order is the same size,
regardless of where volume actually flows that day.

TWAP works well when intraday volume patterns are unreliable (illiquid names, after-hours) or when the trader wants to minimise predictability and information leakage. The cost: timing risk is higher because the trade is spread across the day regardless of where the market actually moves.

Implementation shortfall — the honest benchmark

  IS  =  ( Decision price  −  Average fill price )  ×  ( Filled shares )
         +  ( Decision price  −  Closing price )  ×  ( Unfilled shares )
         +  Commissions and fees

for a buy order. For a sell, swap the signs on the price differences.

where:
  Decision price   =  the market mid-price at the moment the portfolio
                       manager decided to trade — the 'arrival price'
  Average fill     =  the volume-weighted average price actually paid
   price             on the filled portion of the order
  Filled shares    =  the number of shares actually executed before
                       end of window
  Closing price    =  the market price at end of execution window for
                       any shares left unfilled
  Unfilled shares  =  the residual that was not executed
  Commissions      =  brokerage costs, exchange fees, regulatory fees

Implementation shortfall captures the total cost of executing: market impact on the filled portion (the gap between decision price and average fill), opportunity cost on the unfilled portion (the gap between decision price and where the market ended up), plus explicit costs. It is the honest accounting because it includes the cost of what you didn't execute — which traders who only optimise for VWAP can ignore. The IS-minimising algorithm uses real-time market signals to balance impact against opportunity cost dynamically.

Other execution-algo flavours

• POV (Percent of Volume): execute at a defined percentage of market volume (e.g., 10% of every print). Used when filling time is flexible but market impact must be controlled.
• Iceberg: post a small visible quantity at a price; replenish automatically as it fills. Hides true order size from the visible book.

Proprietary trading algorithms

• Market making: post continuous bid-ask quotes, earn spreads. Algos manage inventory, adjust spreads to changing volatility, hedge exposure dynamically. Citadel Securities, Virtu, Susquehanna dominate equity market making.
• Statistical arbitrage: identify pairs or baskets of securities that have historically moved together. When they diverge, bet on convergence. Renaissance Medallion, D.E. Shaw built empires here.
• Latency arbitrage: exploit price discrepancies between venues. If a stock trades 17.50 on NYSE and 17.51 on NASDAQ for a few milliseconds, buy on NYSE and sell on NASDAQ. Requires extreme low latency.
• Trend-following: detect price momentum and ride it. Used by CTAs (commodity trading advisors). Most macro hedge funds use some flavour of this.
• Event arbitrage: trade announcements (earnings, M&A) faster than humans can react. Race to be first to read the press release and trade on it.

What HFT actually is

High-frequency trading is a subset of algorithmic trading characterised by extremely low latency (microseconds), short holding periods (seconds or less), and high transaction volumes (millions of trades per day per firm). Major HFT firms: Citadel Securities, Virtu Financial, Hudson River Trading, Jump Trading, DRW. HFT firms typically run several strategies simultaneously — market making for the bulk of revenue, latency arbitrage as a sideline, statistical arbitrage as a longer-horizon play.

Co-location and the latency race

HFT firms pay exchanges to co-locate their servers in the same data centre as the exchange's matching engine. This reduces the speed-of-light latency from milliseconds to microseconds. Major exchanges have entire buildings of co-located servers next to the matching engine. Specialised microwave networks transmit market data between New York, Chicago, and London faster than fibre — every microsecond of edge matters.

Algo trading in Kenya and East Africa

Kenyan algo trading is very limited. NSE Kenya doesn't have a sophisticated co-location offering. Daily volumes are too thin for HFT to be profitable. Some institutional brokers use simple execution algos (VWAP, TWAP) for large client orders. Prop algorithmic trading is rare. The market structure is closer to 1990s US equity than to 2026 US equity. As exchange technology and volumes grow, algo trading should follow — but it's a 5-10-year journey, not next year.