Tops, shallow dips and bad news: what the new scenarios are built on

2026-10-30 · Strategy

Over the last few days the strategy got four new scenarios and one new filter. Each started with a question anyone looking at a chart asks: can you short the tops? Why does the bot short "at the bottom"? Could we react to a headline before the market digests it? The answers were rarely the ones we expected, which is why every change went through a full backtest on three paths with shifted entry times.

1. Tops look textbook, but shorting them does not pay

We marked every clear swing top and bottom on BTC since 2018 and on SOL and ETH since 2025, and looked at the indicators around them. At tops, price vs the 7-day average, Bollinger %B, CCI and daily momentum always sit at the top of their range. The trouble is that the same values recur all through an uptrend. Shorting "overbought" loses more than it makes on all three coins.

Instead of hunting for the perfect top, we looked for sets of conditions after which a move in one direction was most often worth the risk - consistently, year by year, separately before the ETF era, in it, and on altcoins. Of a dozen or so candidates, three survived the backtest:

The long works on both coins: a calm trend without a wave of forced selling is a good time to enter. The shorts depend on the coin. A negative Coinbase premium says something about US demand for bitcoin, not for Solana. Bitcoins flowing to exchanges while ETF funds sell led SOL lower but hurt on BTC itself. So each coin got its own set. Together: BTC +21% to +48% in the ETF era, SOL +25% to +127%, with the drawdown practically unchanged. The thresholds were picked on the same data, so we expect less live.

2. "Too low to short" - what matters is how deep the fall is, not where the price sits in a window

Breakout Down SHORT once shorted SOL right at the lower Bollinger band, and it looked like a short "at the bottom". We went through all its trades and checked where the price was at entry: against the high and low of the last day, three days and seven days.

The result ran against intuition. The scenario's biggest wins - +62% in January 2026, +53% in May 2022, +37% in December 2021 - entered right at the 7-day low. "Cheap in the window" did not separate good entries from bad ones. The depth of the fall did. Winning shorts came when the price was already 3-7% under its 7-day average; stop losses and zero exits came on shallow dips, 1-2% under it.

• The filter "price at least 2% under the 7-day average" cuts 4-5 of the dozen-odd entries on each path. On BTC before the ETF era the result improves by 5-20%, in the ETF era it changes by 0 to −1%, and on SOL it does not change at all.
• A stricter 3% threshold cut SOL's best short (+35%), which entered 2% under the average.

The filter is not infallible. The day after it went live, a shallow SOL short it would have stopped happened to make money. Over dozens of entries, though, the difference is on the filter's side.

3. Bad news as a signal - and what the Binance hack showed

Since 13 October a language model has been reading the headlines of eight feeds - four crypto, four general - and rating each one: bullish or bearish and how strong, from 1 to 5. By 30 October we had 2,591 headlines. After strong bad news the price usually kept falling:

The night of 29 October was the most instructive. Bitcoin fell from $80.8k to $78.2k between 20:00 and 20:45 UTC. The first headlines about a suspected Binance hack reached the RSS feeds only at 21:12 UTC. The first leg could not have been caught from the news: the market was faster than the newsrooms. But a second leg followed the headlines: down to $76.7k by 00:30 UTC, another 3.5%.

That became News Shock SHORT: a short within an hour of a bad headline rated 4-5, when price momentum is down. To get the news in minutes rather than hours, we added a fast reader that rates only new headlines every 5 minutes. In a backtest on BTC from 13 to 29 October the scenario improved the result by 3.6-5.8% on three paths. That came from just two trades: a short after the PPI inflation shock on 15 October (+2.8%) and a short after the Binance hack (+4.4%). This is an experiment, not proof. We will judge it again once dozens of strong events have built up.

There is no counterpart for good news yet. In 17 days the model rated none at 4 or 5, and weaker bullish headlines said nothing about the price.

4. Two ideas that did not survive

• Closing a position when a scenario of the other side would open. It sounds sensible: the market turns, so we get out. In practice such signals show up on brief pullbacks inside a good move. There were 130-175 such exits per path, and they cut gains the trailing stop would have carried further. SOL lost 25-58%, BTC 15-49% in the ETF era, and the drawdown rose by 7-8 points.
• "Don't short after a panic, it will bounce." We checked Breakout Down SHORT's entries against the previous day's long liquidations. The only time the scenario entered after a day of mass liquidations, in January 2026, it was its best trade ever: +63%. All its stop losses fell on calm days. The panics after Luna and FTX did not end in a quick bounce either.

Summary

• "Overbought" readings are extreme at every top, but also all through an uptrend; shorting them does not pay.
• The new long (a calm daily trend without a liquidation wave) works on BTC and SOL. The shorts depend on the coin, so each coin has its own.
• Breakout Down SHORT makes money after a deep fall; the 7-day average filter now cuts the shallow dips.
• Strong bad news led to further falls, though the market usually prices the first leg before the headline exists. News Shock SHORT runs as an experiment on the fast reader.
• Exiting on an opposite signal and avoiding shorts after a panic looked sensible and cost a lot in the backtest.

The common lesson: a chart suggests stories that sound right - a top, a bottom, a panic, a reversal. Only the list of trades shows which of them is true.