IPO Entry Timing

The recommended order plans — 1 to 4 orders

Each card is a ready-to-use plan. The picture shows where the bargain orders sit below the first traded price (the blue line at 0%). Every order has a deadline — if the price never dips that far, the order turns into a normal buy at the deadline. The four numbers tell you how the plan performed on the past IPOs in the selected window/filter.

When is the cheapest moment to buy?

When the lowest price happened

How big the discount gets

How far below the first price each IPO eventually dropped. Most dip a few percent; a long tail drops 15% or more.

Timing vs size of the discount — one dot per IPO

Dots far right/low: the longer the cheapest moment takes to arrive, the bigger the discount tends to be. Quick dips (left) are shallow.

If you just bought everything at one fixed time…

…this is how much above the eventual low you would have paid, depending on when you buy. The worst time to buy is the first ~20 minutes of trading. This is why the bargain-order plans above beat simply "buying at 10 o'clock".

Which buying method gets the best price?

All numbers = how much above the lowest possible price each method paid. Lower is better. The recommended ladder wins; the machine-learning model and the auto-optimizer don't beat it.

"Buy at preset times" = the best fixed clock times. "Auto-optimizer" rows show what happens if you let an algorithm re-tune the orders for the selected company-size filter — shown only as a sanity check, because that tends to fool itself on small samples. "ML model" = a machine-learning classifier that tries to spot good moments minute by minute.

Does it work on IPOs it has never seen? the honest test

How this was tested: each strategy only ever saw IPOs from the past and was then judged on the 79 newest IPOs it had never seen. Lower numbers = bought closer to the low.

Static ladder — the recommended recipe from above, fixed once and never re-tuned. It beats every strategy that kept re-optimizing itself.
Adaptive policy — watches the first 30 minutes, then either buys immediately or places deep limit orders. Slightly better on day 1, especially in bad cases.
Don't expect much below ~7%. Even a strategy allowed to cheat and look at the test IPOs in advance only reaches 6.7%, and the average dollar actually traded in the market paid 7.8% above the low. The recommended ladder buys cheaper than the average market participant — the rest of the gap to the exact low is luck, not method.
The data behind this is checked: all 157 IPOs passed an integrity audit, day-1 prices match an independent second source, and the order-execution logic is covered by unit tests.

Can machine learning do better?

Short answer: no. A model trained on minute-by-minute trading data finds real patterns, but its buying results are worse than the simple bargain-order plan — so the recommendation stays simple.

What the model pays attention to

What a typical IPO looks like first traded price = 0%

The bold blue line is the typical (median) IPO; the shaded bands show where most IPOs traded — between the bands lies the middle half of all IPOs. Typical pattern: a dip below the first price during day 1, then a slow recovery.

Day 1 — minutes after trading starts

Day 1 evening (after-hours)

Day 2 early morning (premarket)

Day 2 — normal trading hours

All IPOs in the data

Every IPO with its first price, biggest discount and when it happened. Click any column header to sort, click a row to see that IPO's full price chart below.

Look at a single IPO

Generated by analysis/*.py · scenario windows, empirical schedules, signal-pool optimization, LightGBM classifier.