How to Tell Whether a Trading Signal Works When the S&P 500 Is Down

A Rising Market Flatters Nearly Everything

When the S&P 500 climbs, almost any long-biased system looks good. That makes the real test the stretch where the index falls or chops sideways. Before trusting any signal, from any source, you need to ask whether its results came from the signal or from simply holding market exposure while prices rose. This article is a way to answer that. It does not make any claim about a particular YMI indicator, because there is no documented performance record in this article to support one.

A Small Down Year Is Not a Bear Market

The index being down about 2% year to date is a mild decline. It is also a single window of three months or so. A short stretch with a handful of signals can show a profit on pure chance. Before drawing a conclusion, ask how many independent signals occurred, whether the window was chosen after seeing the results, and what the costs were. A three-month period selected because it looked good is not a test.

Separate Market Exposure From Signal Quality

A long-only stock signal will usually move with the index to some degree. Beta measures that sensitivity. If a signal has a beta of 0.8 and the index falls 2%, you would expect roughly a 1.6% decline from market exposure alone. Here is a hypothetical case: if the signal returned +3% over the same period, the portion not explained by market exposure is about 3% - (0.8 x -2%) = 4.6 percentage points. That is a calculation, not a result from any real strategy, and with a short sample the noise around it is large. The point is that the useful number is the return after accounting for exposure, not the raw return.

Why Some Signals Hold Up Better Than Others

Momentum and trend-following signals like moving average crosses depend on persistent price moves, and they tend to struggle when markets chop. Signals built on other kinds of price behavior may behave differently. I say 'may' on purpose. Whether a given structural pattern is less regime-dependent is a hypothesis you test with data, not a property of a design philosophy. The same applies to fixed reference prices like KPL: they are not predictions, and no regime claim should be made without your own results.

The Metrics to Check

• Return during declining periods. Define those periods before looking at results, for example months when the index fell more than a set amount.
• Sharpe ratio. Average excess return divided by its standard deviation, annualized. With only 12 monthly observations, the standard error of an annualized Sharpe estimate is on the order of 1.0, so a reading of 1.2 over a year is weak evidence.
• Maximum drawdown and recovery. The math is lopsided, as the table shows.
• Hit rate across regimes. Compare the win rate in up, down, and flat markets. A strategy that works only in one is a regime bet.

Drawdowns do not recover on their own. They are repaired only by future gains, which are not guaranteed.

A Test Procedure

1. Write the signal rules in full before testing, including entry, exit, size, and what happens on gaps.
2. Fix the test periods ahead of time and include at least one stretch of falling prices and one of sideways prices.
3. Deduct commissions, fees, and a conservative slippage estimate. On index futures, one ES tick is $12.50. Stop orders can fill past the stop.
4. Split the data. Develop on one part and run the finished rules once on the held-out part.
5. Record the number of trades in each regime and state how little they prove.
6. Only then compare against a simple benchmark such as holding the index over the same period.

Common Ways Tests Mislead

• Tuning parameters until a down period looks good, which fits the past and often fails on new data.
• Survivorship bias, such as testing only stocks that still exist.
• Choosing the reporting window after the fact.
• Ignoring costs, which hit high-turnover signals hardest.
• Counting a few trades in a down window as proof.

If you want to run the arithmetic on your own trade log, the free edge calculator is a starting point. For how KPL levels are defined, see the KPL trading page.