Heterogeneous agents, the financial crisis and exchange rate predictability

Abstract

Highlights:• we build an empirical heterogeneous agent model for 6 currencies • individual agent forecasts are constructed from DMA framework • our daily out-of-sample R2 relative to RW can be as high as 1.41% and highly significant • our model forecasts yield annualized Sharpe ratios of up to 1.1 and performance fees above 400 basis points • our predictability results break down after February 2009, are strongest after Lehman Brothers collapse  We are grateful to Andrea Vedolin (LSE) for providing the yield curve factor loadings for Japan and Switzerland. We would like to thank AbstractWe construct an empirical heterogeneous agent model which optimally combines forecasts from fundamentalist and chartist agents and evaluates its out-of-sample forecast performance using daily data covering an overall period from January 1999 to June 2014 for six of the most widely traded currencies. We use daily financial data such as level, slope and curvature yield curve factors, equity prices, as well as risk aversion and global trade activity measures in the fundamentalist agent's predictor set to obtain a proxy for the market's view on the state of the macroeconomy. Chartist agents rely upon standard momentum, moving average and relative strength index technical indicators in their predictor set. Individual agent specific forecasts are constructed using a flexible dynamic model averaging framework and are then aggregated into a model combined forecast using a forecast combination regression. We show that our empirical heterogeneous agent model produces statistically significant and sizable forecast improvements over a random walk benchmark, reaching out-of-sample R 2 values of 1.41, 1.07, 0.99 and 0.74 percent at the daily one-step ahead horizon for 4 out of the 6 currencies that we consider. Forecast gains remain significant for horizons up to three-days ahead. The forecast improvements are largely realised before and around the time of the Lehman Brothers collapse. We show further that our model combined forecasts produce economic value to a mean variance investor, yielding annualized Sharpe ratios of around 1.1 and annualized performance fees in excess of 400 basis points.

Measuring the output gap in Switzerland with linear opinion pools

“…These are: 1) Quadratic trend (QT), 2) Hodrick-Prescott filter (HP), 3) forecast-augmented Hodrick-Prescott filter (HPFC), 4) forecast-augmented Christiano-Fitzgerald filter (CF), and 5) forecast-augmented Baxter-King filter (BK). 4 Best model selection has been applied in dynamic time varying coefficient models to predict exchange rate and copper returns in Buncic and Piras (2016) and Buncic and Moretto (2015). 5 We do not report these here to conserve space.…”

Decision Making by Extracting Soft Information From CSR News Report

“…If a pre-warning system on financial crisis detection is trustful and reliable, then top executives can initiate remedial treatment to prevent the crisis from bursting forth, and investors can grasp any profitability situation that may arise and modify their investment strategies to maximize their personnel wealth under anticipated risk exposure scenarios (Geng, Bose, & Chen, 2015). Thus, detecting early signs of financial vulnerability turns out to be an increasingly urgent task (Buncic & Piras, 2016).…”

Superforecasting: The Art and Science of Prediction. By Philip Tetlock and Dan Gardner

“…In an econometric (model based) forecasting world, this means that one needs to move away from a single model (or predictor) framework, to one that utilizes instead a potentially large number of (small) models, which are then combined or averaged to produce a single forecast of the target variable of interest. Such a modelling strategy is fairly common today (see for instance [3], for an application of this approach in the equity premium forecasting literature, or [4,5] in the context of exchange rate forecasting and equilibrium credit modelling).…”

“…Since this is a critical review, I should say a few words about what I did not like so much in the book. All assessments of forecast accuracy and comparisons across forecasters in the book (and the GJP as a whole) are based on the Brier Score (BS) 4 . Evidently, when wanting to assess how well forecasters are doing, one will need to settle on a performance measure.…”

“…where N denotes the number of predictions that are made,p i is the predicted probability of event i, and y i is the realisation or outcome of the event, where y i takes the value of 1 if the event occurred, and 4 BS seems to be the commonly used abbreviation for the Brier Score (see for instance https://en.wikipedia.org/wiki/Brier_ score). I am thus using it here in accordance with standard practice in the literature, rather than as an expression of contempt or dislike with the measure.…”