Friday, September 20, 2019

Finding Unique Trading Opportunities

Now that I have finished the new book, it's time for a very different project.

The most recent Forbes article outlines a few of the reasons I believe we may be headed for meaningful opportunity in the stock market in the fourth quarter of this year.

Supporting that view is a method for identifying trading opportunities that looks at markets in a unique way.

First, we're looking for opportunity across time frames:  everything from intraday to multiple weeks.  It turns out that traders and investors typically lock themselves into opportunities on a limited range of time frames.  That is very different from asking the question of which time frames, for the current market, offer the greatest opportunity and how can we trade those.

In other words, we're trading where markets offer opportunity, not where our preferences lead us to trade.

Next, we don't look for universal "setups" that will provide trading signals across markets.  Rather, we will break markets into meaningful categories based upon features that differentiate one kind of market from another.  We can think of these categories as "regimes", so that we can categorize rising markets and falling markets; busy and slow markets; markets in high and low interest rate environments; etc.  

This means that, instead of trading the patterns *we* prefer to trade, we identify the patterns that appear in different market conditions/regimes and trade those.

The bottom line is that sometimes we're meant to be shorter-term traders and sometimes longer-term investors.  Sometimes we're meant to trade momentum patterns and sometimes we'll seek value.

The categorization method I'm using is related to k-NN modeling, where we can both classify market types and create regression-based models specific to the classes we identify.  It's important to not get intimidated by the math:  all we're doing is finding market dimensions that matter and identifying the k number of "nearest neighbors" that represent past occasions similar to the current one.  Forecasts are based upon these nearest neighbors only, not upon all days in a backtest.

Backtesting historical periods is not helpful, because we're comparing markets under unlike conditions.  Meaningful patterns in the data are lost if we average all days together.

Interestingly, in the first model I've created (and this is a work in progress, to be sure), we find that markets trading in narrow ranges following strong rises tend to follow through with continued strength over a next 20-30 day period.  That is a similar finding as the one outlined in the Forbes article.

The important takeaway is that we can find value in looking at markets in new ways.  Doing what everyone else is doing and looking at the same things they are looking at is unlikely to offer any meaningful edge over time.  The great problem for many traders is that they keep doing the same things even after markets change.  There are ways of identifying these changes in real time--and the opportunities that spring from them--and I hope to share more as the project unfolds.

Further Reading: