How Quantitative Models Weathered This Week's Drop in US Stocks

On September 29, with two more trading days left in the month, I wrote the following message in the channel: Reviewing the monthly K-line, October is a window for market trend reversal, and the index may begin to show weakness on the monthly K-line. At that time, the model also provided evidence— the X Model started a batch of buy signals again, while high-quality stocks like UnitedHealth and Costco were gradually being bought.

Half a month has passed, and the Nasdaq 100 ETF (QQQ) fell from 759.66 on October 6 to 709.80 on October 16, a decline of 6.56%.

But as I stated then, this assessment was based only on personal experience, K-line, and cyclical interpretation— not as a trading basis. Both the community and the independent site use a full set of model-based trading; the model does not care what I think. The real thing worth reviewing this week is what the model did during the decline.

What happened to the model this week

This week, the model's net value retraced by 5.54%, its biggest single-week drop this year. In the same period, QQQ fell 5.52%, and S&P 500 ETF (SPY) fell 1.97%.

Throughout the week, the model kept buying, nearly all using the X Model.

The X Model is a drawdown model. When a stock falls by a certain margin from its stage high, it begins to buy in batches— the more it drops, the more it buys; when it rebounds to the target price, it sells the entire position. Therefore, in the first four days when QQQ was already down more than 3%, the model kept buying, and by Thursday's close the retracement in net value was less than 0.5%.

On Friday, QQQ dropped another 2.5%, semiconductors saw deeper declines, and the positions accumulated in the previous days took the full hit, causing a single-day net value drop of 5.18%, which accounted for over 90% of the week's retracement.

If you had been following all the model's trades, by Friday's close, the net value would have returned to the level around September 15— an effort equivalent to a lost month.  

8 Trades Completed This Week

During the same week of declines, the model was also selling. This week, 8 trades were completed, all achieving gains, contributing a total of +2.36% to the account.

UNH position is 10%, others are 5% each. This week, 11 new positions were also opened.

The realized +2.36% could not offset the floating losses, so weekly net value still dropped by 5.54%. But this table shows the model is not only buying in a falling market— it is also selling.

Comparing to July's Correction

There was also a round of declines in tech stocks in July. QQQ dropped 8.79% from July 10 to July 29, split into two segments, with a rebound in between.

The first segment of the July drawdown hit chips; QQQ itself did not drop much, and on July 15, QQQ fell only 0.27%, but the model fell 3.65% because the X Model took full positions in semiconductors.

There was a rebound from July 17 to 21, during which QQQ rose 1.96% and the model rose 3.54%. The X Model sold a batch in its entirety during the rebound. By the deepest part of the second segment, the model only fell by half as much as QQQ.

This week is equivalent to compressing both July segments together: QQQ's decline is close to July's second leg, but the model's position structure is like the first leg. In July, the retracement was spread over a few days with breathing room in-between; this time it was four calm days followed by one sudden plunge. In terms of magnitude, this week's 5.54% is less than July's 7.32%,but the pain mainly comes from the speed and intensity.

What's Different Between This Round and July

In July, it was mainly tech stocks that dropped. This time, non-tech stocks fell first, and the model bought into them during their declines. Recently, as tech stocks started falling, non-tech stocks successively rebounded and were sold, and the X Model passively picked up tech stocks— with more frequent rotations.

Currently, most of X Model’s positions are in tech stocks, but it still holds some non-tech stocks.

If You Thought Decline Was Coming at the End of September, Why Not Move to Cash?

If I were trading subjectively, I would control my position in October and plan to bottom fish during the decline. People trading K-line would probably do the same.

But you never know how the market will move tomorrow. If you could be sure of the date and magnitude of a decline, you simply wouldn’t participate. The X Model takes several exploratory entries when an outcome is unknown.

X Model’s design is to add more as it drops, so the fastest position increases actually come when the market is dropping the quickest. This is the cost of the rule itself.

These days, people are again talking about QQQ falling back to the end of July to make a double bottom, or a double top, etc. I look at these technical patterns too, but only as dinner conversation. You can only know a double top/bottom after it’s formed; you can prepare in advance, but trading on a specific pattern isn’t the point of model design.

What the model aims to solve is people’s fear and hesitancy to add positions on a downtrend, and unwillingness to sell when rising. How to handle the market in between is left to the rules. The model as a whole is a quantitative long strategy, so it does not short the market.

This Year's Significant Drawdowns

As of October 16 close, the model's return this year is +49.27%, and SPY is +11.93% over the same period. The notable drawdowns the model has experienced during this timeframe: