dark pool levels

Before reaching out to the dark pool endpoint(s), let's begin with an interesting ticker. Institutional ownership is always a hot topic, so let's see what the Point72 13F filing gives us. We can access 13F data like this:

https://api.unusualwhales.com/docs#/operations/PublicApi.InstitutionController.holdings

We will sort the results by value descending to get a look at the largest positions in the portfolio at the top:

Ticker SHEL jumps out as an Energy name amongst technology and communication firms, let's single that one out.

There are two dark pool endpoints available from UW: one for recent dark pool trades, and one for ticker-specific dark pool trades. Let's start with the ticker-specific endpoint in this example:

https://api.unusualwhales.com/docs#/operations/PublicApi.DarkpoolController.darkpool_ticker

The first thing we will do is hit the endpoint for some data to see if anything jumps out using Shell ticker SHEL:

Nice! rsp contains the API response in json format, so rsp.json()['data'] will give us a list of dictionaries like this:

{'canceled': False,
  'executed_at': '2024-12-06T22:06:46Z',
  'ext_hour_sold_codes': 'extended_hours_trade',
  'market_center': 'L',
  'nbbo_ask': '68.71',
  'nbbo_ask_quantity': 100,
  'nbbo_bid': '63.21',
  'nbbo_bid_quantity': 18,
  'premium': '1511485.25',
  'price': '63.25',
  'sale_cond_codes': 'prio_reference_price',
  'size': 23897,
  'ticker': 'SHEL',
  'tracking_id': 61606319088292,
  'trade_code': None,
  'trade_settlement': None,
  'volume': 5010543},
  ...
{'canceled': False,
  'executed_at': '2024-12-06T14:45:12Z',
  'ext_hour_sold_codes': None,
  'market_center': 'L',
  'nbbo_ask': '63.62',
  'nbbo_ask_quantity': 100,
  'nbbo_bid': '63.61',
  'nbbo_bid_quantity': 1185,
  'premium': '292652.00',
  'price': '63.62',
  'sale_cond_codes': None,
  'size': 4600,
  'ticker': 'SHEL',
  'tracking_id': 35112369931490,
  'trade_code': None,
  'trade_settlement': None,
  'volume': 571115}]

Which is a good start, but let's compile this data into a polars DataFrame to make it easier to manage and analyze:

Let's do some basic data cleanup: then let's add a new column, executed_at_tz, which we will set to eastern time (since the source data is based on UTC):

• executed_at from a string to a datetime
• nbbo_ask from a string to a float64
• nbbo_bid from a string to a float64
• price from a string to a float64
• premium (which is the dollar amount transacted) from a string to a float64

OK great, let's get a quick describe() going here on the most interesting columns, namely:

• size
• price
• premium (a combination, since it is size * price)

Interesting, seeing that 100_000 size as the max makes me wonder if there were lots of transactions at a round 100K share size:

Hmm, so these transactions have Sold Condition Code of average_price_trade (which means the transactions may or may not have actually occurred at this price, since it reflects an average price over a period of time) and Trade Code of derivative_priced (which casts further doubt since it clearly states the material terms of the transaction were not determinable)... Let's cut these out of our dataframe, since we cannot say for sure that a trade actually occurred at those prices.

(Note: in polars I specifically need to allow null values to stick around when filtering this way, which is why you're seeing that is_null() check below.)

Now that our dataframe has been scrubbed, let's repeat these steps and collect dark pool data for lots of trading days, not just today.

Note: the sale_cond_codes, trade_code, and trade_settlement columns will intermittently have string values mixed with nulls and other times will be entirely nulls, which can cause combination issues. We will preemptively cast those columns to strings in order to avoid this hassle.

Awesome, now we have a dictionary with dates (strings) as keys and polars DataFrames as values, let's repeat the data scrubbing steps:

Finally, let's aggregate this "2024 dark pool" polars DataFrame by price to identify the levels at which the most money changed hands (as these might be interesting inflection points):

Wow! I don't know the stats offhand but suffice it to say I would not have expected to see 7 of the 10 biggest dark pool trades in SHEL all happen in the month of April! If I were going to do some more research into SHEL I would definitely start in that time period... Let's take a quick look.

A little "inside baseball" regarding the UW API -- when I need quick closing prices for a security I actually reach for the Realized Volatility endpoint instead of the OHLC endpoint, since the response gives me the daily closing prices ONLY as well as the 30 calendar day implied volatility and 20 trading day realized volatility, two data points I find myself using almsot all the time.

https://api.unusualwhales.com/docs#/operations/PublicApi.TickerController.realized_volatility

Very interesting to see that SHEL had such a steep rally into mid-April before market participants changed their stance, and even though modestly higher highs were observed in May it is clear to see that the steep price appreciation observed from March to mid-April ran out of gas there and since then has come all the way back to price levels last observed in March... Perhaps clusters of large dark pool transactions can signal inflection points in momentum or price?

Let's plot those 10 largest dark pool transactions onto our price chart to confirm or deny this suspicion:

Sure looks like confirmation. Perhaps on a larger (multi-year) scale these "Top 10" trades are not really that big, but in the context of 2024 it is visually clear that these trades established a price "level" in April that became a turning point.