Market Microstructure Analysis

Depth metrics:
  Level 1 depth: queue size at the best bid and best ask
  Level 5 depth: total queue size across the first 5 levels
  Depth asymmetry: (Bid depth - Ask depth) / (Bid depth + Ask depth)
    > 0 -> stronger bid side, price tends to rise
    < 0 -> stronger ask side, price tends to fall

Resilience:
  The speed at which the book recovers after a large-order impact
  Fast recovery -> good liquidity, temporary impact
  Slow recovery -> poor liquidity, persistent impact

China A-share LOB characteristics:
  - The shallowest depth is in the 15 minutes before the open (highest information asymmetry)
  - Depth improves from 10:00-10:30 (institutions begin participating)
  - Best depth is from 14:00-14:57 (most intraday information has been digested)
  - During the 14:57-15:00 closing auction, depth changes sharply (late-day grabbing / dumping)

Order-book imbalance signal:
  OIR = (Bid_vol - Ask_vol) / (Bid_vol + Ask_vol)
  Rolling 5-minute OIR > 0.3 -> short-term bullish signal (accuracy about 55-60%)
  Note: in China A-shares, large orders are often rapidly added and canceled (icebergs / spoofing), so OIR signals need filtering

Flash-crash characteristics:
  1. Price drops more than 5% within minutes
  2. Volume first expands, then collapses (liquidity evaporates)
  3. Bid-ask spread widens sharply (market makers pull quotes)
  4. Followed by a V-shaped rebound (not always fully recovered)

Triggers:
  - Large market order + thin liquidity -> punches through multiple levels instantly
  - Stop-loss chain -> initial selloff triggers more stop orders
  - Algo resonance -> multiple trend-following algos sell simultaneously
  - ETF discount arbitrage -> ETF redemption and constituent selling intensify the drop

Preventive measures:
  1. Use limit orders instead of market orders: specify the maximum acceptable price
  2. Monitor VPIN: if VPIN breaks above 0.5 -> stop trading
  3. Liquidity threshold: exclude instruments with Amihud > 10
  4. Spread monitor: if spread widens suddenly to >5x normal -> pause orders
  5. Time avoidance: do not execute large orders in the first 15 minutes after open or the last 5 minutes before close

China A-share flash-crash cases:
  2015 Jun-Jul: thousands of stocks hit limit-down, with VPIN staying elevated
  2020-07-13: Shanghai Composite plunged and then rebounded in a V-shape
  Pattern: liquidity dries up -> limit-down locking (China-specific) -> next-day panic selling

Call-auction strategy:
  9:15-9:20: orders can be entered and canceled, mostly probing quotes (low reference value)
  9:20-9:25: orders can be entered but not canceled, so real intent is revealed
  Signal: after 9:20, buy orders far exceed sell orders -> likely gap-up open

  Execution: place orders at 9:24:50 (last 10 seconds of the call auction)
  Risk: cannot cancel, and the final execution price may deviate from expectation

Closing call auction (14:57-15:00):
  Feature: closing price is decided within 3 minutes, with concentrated institutional rebalancing and index-fund flows
  Signal: closing-auction volume > 10% of the whole day -> institutions are rebalancing

  Strategy application:
  - VWAP algos should finish most of execution before 14:50, leaving a small residual for the close
  - Avoid placing large orders after 14:57 (high price uncertainty)

Block-trade discount signal:
  Discount = (block-trade price - closing price) / closing price
  Discount < -5%: seller is eager to exit -> short-term bearish
  Discount > -2%: traded near market price -> may be turnover rather than reduction

  Buyer identity:
  Well-known institutional seat buys -> positive signal
  Same broker on both sides -> may be wash trading (neutral)

=== Liquidity Diagnosis ===
Instrument: 000858.SZ Wuliangye
Date: 2026-03-28
Average daily traded value: 2.8 billion RMB  Turnover ratio: 0.85%
Amihud: 0.32 (high liquidity)
Effective spread: 0.05% (2.5bp)
Kyle Lambda: 0.003

=== Order-Flow Analysis ===
VPIN: 0.28 (normal)
Order-book imbalance (OIR): +0.12 (mild bid-side bias)
Net large-order buying: +230 million RMB (institutional buying bias)

=== Trading-Cost Estimate ===
Planned trade size: 500,000 shares (about 40 million RMB)
Estimated impact cost: 0.08% (32k RMB)
Commission: 0.025% (10k RMB)
Stamp duty: 0.05% (20k RMB, sell side)
Total one-way transaction cost: about 0.16%

=== Execution Suggestion ===
Recommended strategy: VWAP
Execution window: 10:00-14:50 (avoid the open and the close)
Number of slices: 5-8 (about 60k-100k shares per slice)
Time sensitivity: low (VPIN is normal, no urgency to execute)

pip install pandas numpy scipy