Market Regime Classification

WHAT IT DOES. It reads findings other scientists already filed and classifies ONE REGIME PER ASSET CLASS, crypto and stocks separately, into one of five labels: bull trending, bear trending, sideways accumulation, sideways distribution or transition. The rule is computed in code, stated with every result, and applied in this order. DIRECTION SCORE: three direction inputs per class each give a lean from -1 to +1, read from that class's assets: ATLAS's trend structure, (uptrend - downtrend) / classified assets; ATLAS's market breadth, (assets above - assets below their 50-bar average) / assets measured; ATLAS's momentum scan, (bullish - bearish) / labelled assets. The class's score is 50 + 50 x the average lean of its inputs present, rounded. LABEL: transition when one direction input leans +0.2 or more while another leans -0.2 or less. Otherwise, when ATLAS's trend structure has more of the class's assets in uptrend or downtrend together than sideways, the market is trending: bull trending at a score of 60 or more, bear trending at 40 or less, transition in between. Otherwise the regime is sideways, and its side is the score's when the score is at least 5 points from 50 (accumulation above, distribution below); inside that margin the class's own liquidity and flow readings break the tie. When the flows point against a score's clear side, the score's side stands and the contradiction is stated. CONFIDENCE starts high and steps down for each missing input, a thin deciding margin, such a contradiction and a large share of high-volatility assets. Every result names each input and what it contributed, so the label can be checked against the readings rather than taken on trust.

WHY TWO REGIMES AND NOT ONE. A single label for a run that holds both crypto and equities would average two markets that need not move together, and would let a crypto flow reading decide a stock regime. So no reading from one class enters the other's regime. Keeping them separate, and then stating whether they agree - the same kind of regime with scores on the same side of 50 - or diverge, shows the reader both states and the relationship between them - and two markets in different regimes is information a blended label would have hidden.

WHY AGREEMENT BETWEEN INPUTS IS WEAKER EVIDENCE THAN IT LOOKS. The right statistical frame is aggregation of CORRELATED SIGNALS. Agreement strengthens a conclusion to the extent that each input contributes information the others do not already contain, so its value depends on their covariance, their shared inputs, and whether their errors are correlated. The three direction inputs here all read largely the same price history over the same universe and window, so some of their agreement reflects a shared input rather than independent confirmation. A student should ask, of any synthesis, what its inputs share before treating concurrence as strength.

REGIME AS A CONCEPT. Regime classification asserts that markets occupy persistent states with different behaviour, which is why the same setup is read differently in each. It has a serious statistical literature behind it - Markov-switching models, following Hamilton (1989), estimate the probability that a series is in each state. Note the distinction those models make: FILTERED probabilities use only data available at the time and are genuinely real-time, while SMOOTHED classifications use later observations and are therefore partly hindsight. A rule like this one uses only the readings available at the run, so it is the filtered kind. A 'transition' category is testable if its criteria are specified in advance, as they are here; a vague transition label applied after the fact is not.