Turning Forecast Highs and Lows into Trading Decisions
Summary
The document frames a strategy-design problem: how to turn repeated forecasts of a security’s future high and low prices into decisions to enter, hold, add to, close, or reverse a position. The assumed inputs include forecast uncertainty and the correlation between high- and low-price errors, alongside the current price. Forecasts are refreshed as new observations arrive, so a useful approach would need to specify decisions over time rather than just predict a price target.
The author says the forecasts come from statistical model ensembles and asks about research on optimal single-security trading and standard strategies for comparison. No strategy, test, or empirical result is supplied in the document itself. It therefore serves as a clearly stated research question rather than evidence that any particular trading rule is profitable. It also leaves important implementation choices open, including how forecasts map to position size, how costs and risk affect decisions, and how the forecast distribution is used to compare actions.
Key ideas
- Repeated forecasts of future highs and lows do not by themselves specify when to enter or exit a position.
- The described inputs include forecast uncertainty and the error correlation between high and low predictions.
- A trading rule must define how positions change as each new forecast arrives.
- The document asks for optimal-trading literature but provides no proposed strategy or performance evidence.
- Position sizing, transaction costs, and risk constraints remain unspecified.
Tags
Full text
# 64078 # How can one, from regular predictions of high & low prices of a security, their variances and error covariance, construct a trading strategy? Suppose I have an algorithm that provides a prediction of the daily high and low of some security n periods in the future, a confidence interval for each of these predictions, and the correlation coefficient for the errors of these pairs of predictions. And of course, I have the current price now. Suppose, moreover, that all of these predictions are well-validated and correct – not perfect, but with the imperfection correctly described by the terms given above. Each time a period (whether hour, day, week, or whatever) goes by, I then do a new forecast with an additional period's data, and get all the same estimates. Although that seems like a lot to know, and like more than most investors do know, I am having great difficulty in translating this prediction information into a trading strategy. By a trading strategy here I mean a time path or decision rule that tells me whether I initially go long or short or wait, and then, having established a position, on doing my next estimate, whether to liquidate my position, double up on it, liquidate and go the opposite way, wait and do nothing, or what? I am essentially looking for the strategy that lets me use these forecasts to profit from short-term price movements. It seems like I have learned a lot about forecasting, and about portfolio construction, but almost nothing about this sort of, I suppose one might say pejoratively, speculative trading. Is there a literature that describes optimal trading of a single security given information of this sort? Are there standard strategies in common use, that I could compare and choose between? Note that I am not looking for the sort of strategies that I think of as more typical of technical traders. My forecasts are from ensemble averages of, e.g., penalized cointegrated VECMs, penalized ARIMAX models, and other models of that ilk. I want to construct strategies based my forecasts, not on whether I have bounced off a 20-period moving average.
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.