LessonThis is a lesson restated in new words. Any figure in it is an example from the source, not this desk's record.
A strategy that ignores the regime it is trading in will misread its own backtest statistics. This lecture builds the shell of a live screen: five-second bars from Interactive Brokers tick data, a dark dashboard, and a button that refits a Markov volatility model on the latest 60 bars. The model itself is only a placeholder here. Part two fills in the probabilities. The feed is what makes a regime label refer to the market in front of you rather than to a stale file.
Why it matters
Low, middle, and high volatility, or momentum versus reversion, or a burst of attention, are different conditional worlds. Pooling trades across them mixes distributions. The screen in the lecture is meant to show which world a short-term bar thinks it is in, so a later strategy can condition on that label. It is a classifier sitting on a broker connection, not an order sender. The demonstration uses Apple, paints recent candles, and starts a five-second bar stream. Calibration looks back 60 bars, short enough that the labels are not described as stale. In the demo the labels moved among low, high, and mid. A recalibrate control throws away the old fit and uses the latest 60 bars at that moment.
The idea in plain words
The connection class implements both sides of the Interactive Brokers API. The client sends requests, whether through the gateway or Trader Workstation. The wrapper receives replies. A contract object names the instrument. Equities are the example. Futures or options would use the same socket pattern.
Errors arrive with a request id, a code, and a message. Codes 2104, 2106, 2158, and 2176 are treated as connection or farm-status notes and ignored. Code 10167 means the feed is delayed, which is printed and then skipped. Anything else is printed with its id. The next-valid-order-id callback is used only as proof the socket is up, because this build does not send orders. A print says the workstation accepted the connection.
Historical bars come back one at a time. Each request id gets its own list of open, high, low, and close, so a recalibration cannot reuse yesterday's list by accident. When the historical request ends, a threading event is set. Live ticks follow. A nonpositive price is dropped. Tick type 4 is the last trade, type 1 the bid, type 2 the ask. The last trade is handed to the interface with a timestamp. Volume, via tick size, is left as an extension for a volume regime.
How to use it
The window is a Tkinter dashboard, 1200 by 800, background color 0D1117, using the clam theme. It owns one broker object and a callback for each tick. Streaming starts off. Bar length is five seconds. A fixed queue holds ten completed candles plus the one forming. Recent tick prices sit in another queue, capped at 100. Connecting uses the local host and port you configured. After the link, you pick a symbol, request history long enough to fit the model, wait for the historical-data event, and only then start the live stream. Recalibrate repeats the history request under a new id and refits.
What usually goes wrong
Fitting on a long history makes the regime a description of last month. The lecture explicitly refuses that for a short-term screen. Fitting before the historical callback finishes means the model sees an empty or partial list. Mixing request ids mixes old bars into the new fit. Treating status codes 2104 and its cousins as failures will look like a broken socket on a healthy login. Forgetting to ignore nonpositive prices will write a zero or a negative print into a candle and then into a volatility ratio.
Questions people ask
Why five-second bars and 60-bar refits?
The lecture wants a short horizon. Five seconds builds the candle. Sixty bars is the calibration window, short enough that the speaker does not call the regimes stale. Recalibrate repeats that window from the current time.
Which tick is the traded price?
Tick type 4. Type 1 is the bid and type 2 is the ask. Prices at or below zero are discarded before any of those fields update.
What does the volatility number measure?
In this build, high minus low, divided by the close when the close is positive. The lecture treats that as a simple choice you can replace.