Understanding time horizons in signal trading
The time horizon is the average time between entering and exiting a trade.
Quantitative managers calculate a trading signal according to a specific time horizon ranging from microseconds (high-frequency) or a few hours (intraday); to a few weeks (mid-term) or even a few months (long-term).
The SYGNAL platform offers strategies along a standardized spectrum of time horizons:
|Time horizon||Average trade duration|
|Intraday||Up to 24 hours|
|Intraday to Short-term||Up to 7 days (including intraday)|
|Short-term||1 day to 7 days|
|Short to Mid-term||1 day to 60 days|
|Mid-term||8 days to 60 days|
|Mid to Long-term||8 days to 5 years|
|Long-term||61 days to 5 years|
|Diversified||A mixture of multiple time horizons|
Interpreting time horizons for trading
According to a different time horizon, the same signal model can result in very different trading behavior, each with pros and cons.
Shorter-term time horizons: high-frequency, intraday, short-term
Generally speaking, signal models with a shorter time horizon will trade more often than signal models with a longer time horizon and tend to perform best during extreme conditions such as flash crashes. However, more trading can lead to more “false alerts” (especially in choppy market environments), higher trading costs, and more significant slippage.
Short-term equity strategies generally had terrible years with poor track-records in the QE-driven bull market that began in 2010. However, during the March 2020 market turmoil, shorter-term models outperformed significantly.
For example, compare a simple short-term, trend-following model with a 10-day lookback window to a long-term, trend-following model with a 100-day lookback window, both applied to the S&P 500 Index during the start of the COVID-19 pandemic in Europe and North America.
In March 2020, as the pandemic spread through Europe and North America, the short-term model indicated exiting the market on February 21, 2020, only two days after a previous maximum in the S&P 500 Index. The model would have re-entered the market on March 25, 2020 (after 23 trading days), and, therefore, more easily captured upside trends after the market bottomed on March 23, 2020 (in fact, two trading days later).
The same model with a long-term time horizon would have indicated exiting the market on February 25, 2020 (four days later), and re-enter the market on May 26, 2020 (after 61 trading days), therefore missing out on much of the early market rebound.
The above example might look like perfect market timing for the short-term signal model, and indeed it would have worked well. However, short-term models can cause significant losses in choppy markets and are not necessarily the best choice in all market conditions.
Longer-term time horizons: mid-term, long-term
Longer-term models are better in choppy or sideways, non-directional conditions and do a better job at avoiding “false alerts.” Such models generally incur less trading costs and reduce the risk of slippage.
Let’s adopt the dataset from above and focus on the post-drawdown period from March to October 2020. Although the short-term model performed well when the crisis hit, they did poorly during the market’s recovery, which was a rollercoaster of lows to new all-time highs. Expressly, short-term signal models traded actively in and out of the market, for a staggering total of 30 times (or 15 turns), incurring costs.
By comparison, the 100-day (long-term) model only turned the position four times over the same period. In other words, the longer-term models acted as they were supposed to: weed out the noise and ride the market.
Diversified time horizons
The diversified time horizon spans multiple periods: from short-term to midterm or mid-term to long-term. Typically, the diversified time horizon results from combining various models, each with a different time horizon. So the combination of a mid-term trend-following model with a long-term trend-following model generates a final model, or signal, that has a diversified mid-to-long-term time horizon.
Models with diversified time horizons come in handy when the instrument’s price behavior varies over time due to extraordinary corporate events (e.g., restructuring, insolvency) or events that impact expected corporate growth and profitability (e.g., supply-demand shifts).
There is a large number of variations on how signal managers determine time horizon parameters.
When using short-term strategies, an investor should ensure that the traded asset is sufficiently liquid (to avoid slippage), that the fees are low, and that the market environment is right for the specific signal.
Longer-term strategies generally follow the old economic axiom of “a walk in the woods.” They will guide you over rough terrain and help you to capture better returns. However, keep an eye out for the cliff ahead.
Paying attention to both shorter and longer-term models generally leads to the best results.
Time horizon vs. investment horizon
The Time Horizon is NOT the same as the investment horizon. Traditional financial advisors often define the “time horizon” as the period in which you can invest some savings until a liquidity event, such as purchasing a house, occurs. In the quant world, we call this Investment Horizon. Generally speaking, a longer Investment Horizon allows a trader to take more risks over the period than a short Investment Horizon.
Time horizon vs. holding period
The Time Horizon is NOT the same as the holding period. The holding period is the average amount of time a trade is "on," i.e. when the investment is held between the buy and the sell, as indicated by the signal. Thereby, quant managers measure the holding period in days, weeks, months, or even years. For instance, using long-only signals (0.00 to 1.00), the holding period starts on the day when the signal begins to increase from 0.00, and the trader would open the position. The day the signal turns back to 0.00, the trader would close the position. The holding period is then the difference in time between the buy and sell.