The Minute Subtleties of Trading Holding Times
Today’s “Long Story Short” discusses how holding time affects a lot of the considerations for your strategy. Kollider’s series is purely educational and not investment advice. Please do your own research.
Whether you’re a hobbyist, a trader, or an investor, there’s a question that has and probably continues to stump you. When should you sell your position? Once you get into a trade, how long do you hold on for?
In trading, time cuts through everything. It may sound overly broad, but in this article we’ll see how changing one’s holding time impacts the entire strategy. We’re not going to answer the question of how long a position should be held. Instead, we’ll highlight some of the major factors that your holding time influences.
Before we move on, let’s set the scene.
First and foremost, what do we mean by “holding time”? Specifically, we are referring to the length of time that you might hold onto your positions after entering into a trade. This doesn’t necessarily mean a fixed, constant time every time you get into a trade. For example, you might have a strategy that has trades that stay open anywhere between 30 seconds to, say, 15 minutes. This is quite different from a strategy that readjusts and rebalances the portfolio between 9 and 18 months. One assumption we’re making is that even your event-based signals will still have some cadence embedded in it (e.g., a momentum signal that tracks big surges in volume will still have a half-life).
Second, we’ll be going through concepts fairly quickly, so it will be helpful to review some basics. In particular, the Kollider articles on the minimum viable strategy, algo trading and backtesting, and turning an algorithmic strategy into a yield-generating product could be useful here. Although these don’t directly touch on the topic covered today, they give enough foundation related to today’s topic.
With that out of the way, let’s get into the heart of it.
When you change your holding time, it alters a lot of variables in your strategy pipeline. Let’s illustrate which ones we think are important with a comparison of a longer-term strategy with, say, a holding time of 1–3 weeks and one with just a holding time of 30 seconds to 10 minutes.
- Types of data needed. Shorter-term traders often rely heavily on price data, since price action can be a great source of signal, especially in the realm of seconds through low hours. While a short-term trader could use other inputs, prices are typically sufficient to at least construct a working model. For longer-term trading and investing, it would probably be useful to use additional sets of data like economic numbers, macroeconomic indicators, and sentiment for the products they are trading.
- Price data granularity. With weekly holding times, your strategy will likely use price bars that are at least daily, if not weekly, periods. With the shorter-term strategy, there’s not much value in looking at what happens in weeks and so your price aggregation will probably be in seconds, minutes, and hours.
Usually, the shorter your holding period, the more granular your data. High-frequency trading strategies, for example, often require L2, if not L3, data. However, these are of less use to someone trading for weekly time horizons.
- Shape of the price return histogram. The magnitudes of price returns vary significantly across time frames. In addition, at lower holding times, when a strategy is more likely to have periods of not holding a position, traders may be able to ignore or eliminate anomalous price changes. For instance, if a product’s price moves typically 1–2% per day but sometimes can move 5–10% in a single day, a shorter-term strategy may not notice the 5–10% move or may treat it as an uncommon trading environment and turn off if it’s not performing well. In this case, the effective histogram for the shorter holding time strategy reduces the tails.
Because of this, you might be able to use different models for different time periods. For example, if one histogram starts to look more like a normal distribution, you might opt for, say, a simpler linear regression as your base model.
- P&L volatility, margin requirement, and thresholds for stop loss and take profits. Similarly to price returns, a strategy’s profit and loss volatility will vary based on the holding period.
Longer-term trading results in a lot more fluctuation in unrealised P&L due to positions being open longer. The risk of liquidation (or margin calls) increases when P&L volatility is higher. As a result, you will need to place stop losses and take profits farther out (especially with the higher volatility, placing stop losses too close could make your win-loss ratio much worse).
Similarly, your realised P&L will also be more volatile and your sharpe ratio lower (at least a variation of sharpe ratio that removes the risk-free rate to adapt it to be more of a measure of stability than stability of excess returns). These may be more subtle, but this is partly because of likely-increased sizes that you would take. Due to the fact that longer-term trading will also result in fewer position turnovers (i.e., entering and exiting trades less frequently), you may consider increasing your position sizes (and longer terms will allow you to accumulate more size due to increased liquidity). A good strategy will still have a good P&L on average, but a sampling of a long-term strategy’s performance on a day-to-day basis will vary greatly. Reusing an oldie-but-goodie chart from our previous article on algorithmic trading, the longer-term strategy will see much deeper, unrealised drawdowns.
- Forecast horizon. As a result of the holding times, your forecasting horizon will also vary. For the 30 second to 10 minute holding time, it might not make sense to predict where prices will be 1 week from now. Those prices have minimal impact on your strategy. On the other hand, 15–30 second predictions are of little value for the 1–3 week strategy. Long-term strategies would benefit from forecasting where their positions will be in the near future (e.g., trailing stop-loss and take-profit orders).
- Signals. Due to the change in forecast horizon, the signals that will be most effective for your holding times will also differ. The more obvious examples are directly or indirectly time-based signals (think indicators using moving averages). You will have to adjust the underlying moving average periods used to scale with your holding period. As we mentioned above, event-based ones will probably also have their own time cadence. This means that if there is an event-driven signal that tends to correlate with 0.5% price moves over 30 minutes. A strategy that maintains its positions for weeks may not find much use for this, but one that trades in minutes probably would.
Finally, the resources you need will also depend on your holding times. We didn’t discuss how to figure out your optimal holding time (only you can determine that). In the end, it comes down to how you handle the variables above and what you need to do to support them for the time frame you choose.
We already discussed the increased margin required as your time horizon increases, but a shorter-term strategy may also require increased upfront infrastructure costs. The reason might be ingesting, recording, cleaning, and re-fetching a ton of granular data (e.g., level 2 or level 3) that you may not need for a longer time frame. The ability to consume this data in a timely manner is a problem on its own. The process is often much easier when you don’t care as much about its timeliness as you do about its quality and consistency.
So, what is optimal will strongly depend on the data, technology, skills, time, capital, and investors at your disposal.
Know anyone else interested in learning more about bitcoin, Lightning, or trading? Follow us on Twitter and help us share the word.