Golden Cross

What a golden cross is

A moving average is the average of the last N closing prices, recalculated on every new bar. A bar is one candle of whatever time frame you chose: a minute, an hour, a day. A simple moving average (SMA) gives every one of the N closes the same weight. An exponential one gives recent closes more weight, which is the subject of EMA crossover.

The 50 bar average reacts quickly because it only looks back 50 bars. The 200 bar average reacts slowly because it looks back 200. When the quick line moves from below the slow line to above it, that event is called a golden cross. People read it as a sign that recent prices have climbed far enough to pull the short average over the long one. When the quick line falls back under the slow one, it is called a death cross.

Nothing about the numbers 50 and 200 is special in the arithmetic. They are conventions, chosen because they are round and widely used. On daily bars they cover about ten weeks and about forty weeks of trading. On a five minute chart they cover a few hours and a day or two. The same words, golden cross, are used for all of these, and the meaning changes with the time frame.

The 200 bar average needs 200 closes before it exists at all, so the earliest a cross can happen is bar 201. A bot that starts with a fresh chart has to load at least that much history first.

A worked example with made-up prices

The picture at the top uses a made-up path of 320 closes. It falls from 100 to about 80 over the first 150 bars, climbs to about 120 by bar 260, and then drifts down slowly, with a small wave added so it is not a straight line. Every number here is invented to show the arithmetic, and none of it describes a real market.

For the first 200 bars the 50 bar average sits below the 200 bar average, because the price has been falling. Then the climb begins. Around bar 202 the quick line catches the slow one. The table shows the three bars around the cross.

The gap is SMA 50 minus SMA 200. It is negative before the cross and positive after. The bar where it first turns positive is the cross bar, 202. That is when a rule written as "buy on a golden cross" would act, at about the next price it can get.

How late it fires

Both averages are built from old prices, so the cross can only appear after the price has already moved. In the made-up path the lowest close was 78.67 at bar 146. The cross appeared at bar 202, which is 56 bars later, at a close of 99.37. From the low to the cross the price had already risen 20.70. The whole low-to-high rise in the path is 42.98, so roughly 48 percent of it was behind the cross before the signal existed.

That is the trade-off in plain view. A longer average is steadier and gives fewer false alarms, but it waits longer. A shorter average reacts sooner and gives more false alarms. You cannot have both. A bot that uses the cross as an entry has to be built knowing that part of the move is gone, and that the stop point or exit rule has to be judged against that late price.

The delay also depends on the shape of the move. A sharp V-shaped recovery pulls the 50 bar average up fast, but the 200 bar average barely moves, so the cross comes quickly. A slow grind up takes a long time to lift the quick line over the slow one. The price moves first, the 50 bar average second and the 200 bar average last, so the cross is always the final step in the chain.

Whipsaw: when the lines keep swapping

A trending market is the friendly case. A sideways market is the hard one. When the price swings up and down around a level, the 50 bar average wanders back and forth across the 200 bar average, and each crossing looks like a new golden cross or death cross. Traders call the repeated false starts whipsaw.

The second made-up path has 600 bars and moves sideways with two waves of different length added on top of a tiny drift. A rule that goes long on each golden cross and sells at the next death cross produced 3 trades:

Each entry came after a rise and each exit after a fall, and the exit price was lower than the entry price every time. The total is -27.17 per unit, before any spread or slippage. Those costs would only make it worse. This is one made-up path chosen to show the mechanism. It does not say how often this happens in real markets, and it does not say the rule always loses. It says the rule has no way to tell a trend from a swing until after the fact.

How a bot rule is written

A bot can only act on information it has when the bar closes. A golden cross rule is therefore written in terms of closed bars. In plain steps it looks like this:

• Inputs. The fast length (50), the slow length (200), the type of average (simple here) and the time frame.
• Warm-up. Do nothing until there are at least 200 closed bars.
• On each closed bar. Compute SMA 50 and SMA 200 for this bar and for the previous bar.
• Entry condition. If SMA 50 was at or below SMA 200 on the previous bar and is above it now, and the bot is flat, open a long position on the next bar.
• Exit condition. If SMA 50 crosses back below SMA 200, close the position. Many designs also add a protective stop, which is a separate rule with its own setting. Position size comes from a separate rule too, for example position size.

Two details cause most of the bugs. The first is using the current, still-forming bar. The average on an unfinished bar changes with every tick, so a cross can appear and vanish within the same bar. Use closed bars only. The second is using a price from the future when testing, for example acting at the same close that created the cross. Real orders go in after the close. See look-ahead bias.

How bots break it

• Whipsaw losses. Many small losing trades in a range, each paying the spread. The losses are small one at a time and large in a row. Check the max drawdown across a whole sideways stretch, not only the trends.
• Late entry, late exit. The exit is a death cross, which also fires late, so a bot can give back a large share of a gain before it exits.
• Few signals, fragile statistics. On a daily chart a 50 and 200 cross happens only a handful of times in many years of data. A test with a few trades says very little, and picking the best lengths from a grid of choices is overfitting.
• Different averages, different dates. A simple average and an exponential one cross on different bars. A tool that uses a different type, or a different price such as the high, gives a different date for the same picture.
• Gaps and thin trading. A cross is computed on closes, but an order fills at a real price. If the price jumps between bars, the fill can be far from the close that triggered it.

How to check a rule that uses it

Test it on more than one kind of market: a long uptrend, a long downtrend and a long sideways stretch. Count the trades, and look at the average result per trade after costs and the longest run of losing trades. Then keep part of the data back and test once on it, as in how to backtest. If the rule only works for lengths very close to 50 and 200 and fails at 45 or 55, the result depends on a lucky setting, not on a feature of the market. Compare it also with the plain EMA crossover, breakout trading and doing nothing.

It can also be combined with other tools, such as the RSI indicator, Bollinger bands or VWAP. Each addition is another setting that can be fitted to the past, so the same tests apply to the whole bundle. A rule that fires on a cross and then fails at once is a close cousin of a false breakout.

What it does not tell you

It does not say where the price goes next. It says that, over the last 50 bars, the average close is higher than the average over the last 200. That is all. It does not say how long a move will last, how far it will go, or whether it will reverse. No setting of it removes risk, and this page is education, not advice.