Forecast Horizon Increases Research Articles

Application of AI for Short-Term PV Generation Forecast.

The efficient use of the photovoltaic power requires a good estimation of the PV generation. That is why the use of good techniques for forecast is necessary. In this research paper, Long Short-Term Memory, Bidirectional Long Short-Term Memory and the Temporal convolutional network are studied in depth to forecast the photovoltaic power, voltage and efficiency of a 1320 Wp amorphous plant installed in the Technology Support Centre in the University Rey Juan Carlos, Madrid (Spain). The accuracy of these techniques are compared using experimental data along one year, applying 1 timestep or 15 min and 96 step times or 24 h, showing that TCN exhibits outstanding performance, compared with the two other techniques. For instance, it presents better results in all forecast variables and both forecast horizons, achieving an overall Mean Squared Error (MSE) of 0.0024 for 15 min forecasts and 0.0058 for 24 h forecasts. In addition, the sensitivity analyses for the TCN technique is performed and shows that the accuracy is reduced as the forecast horizon increases and that the 6 months of dataset is sufficient to obtain an adequate result with an MSE value of 0.0080 and a coefficient of determination of 0.90 in the worst scenarios (24 h of forecast).

Time interval choices in forecasting stock market indices of CEE and SEE countries

ABSTRACT The main objective of this analysis is to investigate how varying the forecast horizon and the input window length for calculating technical indicators affects the predictive performance of different machine learning algorithms on forecasting the direction of change of chosen stock market indices. Ten indices from CEE (Central and Eastern European) and SEE (Southern and Eastern European) countries are chosen for research in an attempt to investigate their behaviour in the light of the behaviour of bigger and more researched markets. In respect to similar research conducted on S&P 500 Index stocks, this analysis does not find the same pattern of highest system performance for each forecast horizon value when the input window length is approximately equal to the forecasting horizon. Instead, the forecasts seem to be better using shorter input window lengths for technical indicators in general. Also, on average, there is a notable deterioration in the performance with the increase of forecasting horizon. Furthermore, some algorithms perform very well for short horizons and then deteriorate substantially as the forecasting horizon increases, while others seem to have more consistent performance over different horizons.

Adjusting to Rain Before It Falls

Unchecked climate change will cause precipitation volatility to increase around the world, leading to economic damages in the face of adjustment costs. We estimate these damages for construction—an economically important, climate exposed industry. Empirically, employment falls in response to forecasted rainfall and more so as the forecast horizon increases. This pattern allows for identification of labor adjustment costs via a multisector model of local labor markets calibrated to our estimates. When rainfall is anticipatable one month ahead, construction firms pay 10% of monthly profit to adjust. They pay less than 1% for rainfall anticipatable six months ahead. Without further adaptation or forecast improvements, increased rainfall volatility due to climate change is projected to lead to more costly adjustment. This paper was accepted by Elke Weber, Special Section of Management Science on Business and Climate Change. Supplemental Material: The data files and online appendices are available at https://doi.org/10.1287/mnsc.2023.4697 .

A Hawkes Model Approach to Modeling Price Spikes in the Japanese Electricity Market

The Japan Electric Power Exchange (JEPX) provides a platform for the trading of electric energy in a manner similar to more traditional financial markets. As the number of market agents increase, there is an increasing need for effective price-forecasting models. Electricity price data are observed to exhibit periods of relatively stable, i.e., low-magnitude, low-variance prices interspersed with periods of higher prices accompanied by larger uncertainty. The price data time series therefore exhibits a temporal non-stationarity characteristic that is difficult to capture with typical time series modeling frameworks. In this paper, we implement models for the occurrence of price spike events where spikes are defined as observing prices above a predefined threshold set here at 25 JPY/kWh. This value corresponds to about the 90th percentile of observed prices during peak trading periods. The price spikes time series is observed to be rare events that occur in clusters. We therefore propose to model the data as a Hawkes process whereby the occurrence of a spike event increases the probability of observing more spikes in the period immediately following a price spike event. We test two variations of the classical Hawkes model: the first variation models the change in the magnitude of the underlying intensity as a function of the magnitude of the price spike while the second variation models the change in the decay rate of the underlying intensity as a function of the magnitude of the price spike. An analysis of the performance of the models based on the mean absolute error (MAE) of the spike occurrence probability, a weighted accuracy index, and the Matthews correlation coefficient (MCC) metrics shows the effectiveness of the variable magnitude variation of the Hawkes model in generating short-term forecasts of the occurrence of price spike events. The modified Hawkes model especially outperforms other candidate models as the length of the forecasting horizon increases.

Forecasting a Commodity-Exporting Small Open Developing Economy Using DSGE and DSGE-BVAR

In this paper, we assess the forecasting performance of three types of structural models – DSGE, BVAR with Minnesota priors, and DSGE-BVAR – in the context of a commodity-exporting small open developing economy using the data for Kazakhstan. We find that BVAR and DSGE-BVAR models generally produce point forecasts that are more accurate and less biased compared to those of DSGE in the short term, but that BVAR forecasts rapidly deteriorate in quality as the length of the forecast horizon increases. The density forecast analysis shows that when all variables are jointly considered, the models have similar prediction accuracy, and when financial sector variables are omitted, the BVAR and DSGE-BVAR models demonstrate superior performance in the short term.

Long-Term Streamflow Prediction Using Hybrid SVR-ANN Based on Bayesian Model Averaging

The Volga River, as the primary supplier of the Caspian Sea, plays a huge role in its ecosystem sustainability. In this study, we analyze its runoff predictability for different monthly forecast horizons. Additionally, the meteorological and hydrological variables affecting runoff in each month are identified. A wide range of the potential variables was first collected and the Boruta variable preprocessing method was employed to select the important ones. Then the hybrid models were created by combining the selected variables and the data-driven models [i.e., support vector regression (SVR), artificial neural network (ANN), and multiple linear regression (MLR)]. To postprocess the predicted data, the Bayesian model averaging (BMA) method was employed using the combination of the Boruta–artificial neural network (B-ANN) and the Boruta–support vector regression (B-SVR) models. Finally, the Kling-Gupta efficiency (KGE) and the continuous ranked probability skill score (CRPSS) probabilistic evaluation criteria were applied to evaluate the hybrid models. The results showed that the streamflow of the previous steps is the most crucial variable in predicting the streamflow of all next horizons, while its significance decreases as the forecast time horizon increases. Moreover, the temperature variables have the unlike effect on the streamflow prediction and the minimum temperature for winter and spring and the maximum and the average temperatures for summer and autumn are the most effective ones. Correspondingly, BMA-B-ANN-SVR presents the best performance among the hybrid models (for example, the median of 0.8 and 0.91 for CRPSS and KGE in the first horizon, respectively); the reliability of its predicted runoff for different forecast horizons is much better than other hybrid models (B-ANN and B-SVR).

Out-of-sample forecasting of foreign exchange rates: The band spectral regression and LASSO

We propose to utilize the band spectral regression for out-of-sample forecasts of exchange rates. When one period ahead forecast is considered, there is some evidence that the band spectral regression improves its accuracy, especially when the Taylor rule fundamentals model is employed. However, when the forecasting horizon increases, the purchasing power parity (PPP) fundamentals model is found to be powerful, and we can improve the out-of-sample forecast by selecting appropriate frequency bands. Bayesian model averaging shows that placing a large weight on the business cycle frequency improves the accuracy of the out-of-sample forecasting of the PPP model (as well as the monetary fundamentals model) when a longer forecasting horizon is our focus. Without specifying the frequency bands prior to applying the regression, LASSO can provide better out-of-sample exchange rate forecasts for many cases – most patently for the PPP fundamentals model – and provide information about the dynamic relationship between forecasting variables and exchange rates. The frequency domain approach not only improves the accuracy of exchange rate forecast but provides insights for understanding why the PPP fundamentals act as a powerful predictor when the forecasting horizon increases and there is a possible improvement in the time domain regression forecast.

Multi-objective regression modeling for natural gas prediction with ridge regression and CMARS

Residential customers are the main users generally need a great quantity of natural gas in distribution systems, especially, in the wintry weather season since it is particularly consumed for cooking and space heating. Hence, it ought to be non-interruptible. Since distribution systems have a restricted ability for supply, reasonable planning and prediction through the whole year, especially in winter seasons, have emerged as vital. The Ridge Regression (RR) is formulated mainly to decrease collinearity results through shrinking the regression coefficients and reducing the impact in the model of variables. Conic multivariate adaptive regression splines ((C)MARS) model is constructed as an effective choice for MARS by using inverse problems, statistical learning, and multi-objective optimization theories. In this approach, the model complexity is penalized in the structure of RR and it is constructed a relaxation by utilizing continuous optimization, called Conic Quadratic Programming (CQP). In this study, CMARS and RR are applied to obtain forecasts of residential natural gas demand for local distribution companies (LDCs) that require short-term forecasts, and the model performances are compared by using some criteria. Here, our analysis shows that CMARS models outperform RR models. For one-day-ahead forecasts, CMARS yields a MAPE of about 4.8%, while the same value under RR reaches 8.5%. As the forecast horizon increases, it can be seen that the performance of the methods becomes worse, and for a forecast one week ahead, the MAPE values for CMARS and RR are 9.9% and 18.3%, respectively.

Forecasting the volatility of agricultural commodity futures: The role of co‐volatility and oil volatility

AbstractWe forecast the realized volatilities of China's agricultural commodity futures (corn, cotton, palm, wheat, and soybean) using a set of multivariate heterogeneous autoregressive (MHAR) models. We consider different error structures to capture the co‐movement of volatility (co‐volatility) between commodity futures to obtain out‐of‐sample forecasts of the realized volatilities of agricultural commodity futures at daily, weekly, and monthly forecast horizons. We also consider global oil volatility as an additional exogenous predictor and assess forecast precision based on both statistical and economic value measures. The results show that the MHAR model with a co‐volatility error structure is superior in predicting the volatility of commodity futures. The most accurate predictions are of corn, cotton, and wheat futures volatility. Interestingly, for all the commodities, the forecasting accuracy noticeably improves as the forecast horizon increases. However, from an investment perspective, the highest economic value is gained from medium‐horizon forecasts, and the economic value of forecasts is highest for less risk‐averse investors.

Can the Baidu Index predict realized volatility in the Chinese stock market?

This paper incorporates the Baidu Index into various heterogeneous autoregressive type time series models and shows that the Baidu Index is a superior predictor of realized volatility in the SSE 50 Index. Furthermore, the predictability of the Baidu Index is found to rise as the forecasting horizon increases. We also find that continuous components enhance predictive power across all horizons, but that increases are only sustained in the short and medium terms, as the long-term impact on volatility is less persistent. Our findings should be expected to influence investors interested in constructing trading strategies based on realized volatility.

Stock market volatility forecasting: Do we need high-frequency data?

The general consensus in the volatility forecasting literature is that high-frequency volatility models outperform low-frequency volatility models. However, such a conclusion is reached when low-frequency volatility models are estimated from daily returns. Instead, we study this question considering daily, low-frequency volatility estimators based on open, high, low, and close daily prices. Our data sample consists of 18 stock market indices. We find that high-frequency volatility models tend to outperform low-frequency volatility models only for short-term forecasts. As the forecast horizon increases (up to one month), the difference in forecast accuracy becomes statistically indistinguishable for most market indices. To evaluate the practical implications of our results, we study a simple asset allocation problem. The results reveal that asset allocation based on high-frequency volatility model forecasts does not outperform asset allocation based on low-frequency volatility model forecasts.

Multi-step-ahead forecast of reservoir water availability with improved quantum-based GWO coupled with the AI-based LSSVM model

It remains a challenge to obtain an accurate multi-step-ahead forecast of reservoir water availability for island areas. A novel hybrid multi-step-ahead forecast model is introduced to solve this problem. Partial autocorrelation function (PACF) is first used to analyze the characteristics of the target time series for extracting the appropriate input variables. Least-square support vector machine (LSSVM) with recursive mechanism is then explored for modelling multi-step-ahead forecasts, in which the forecast model is adjusted adaptively as long as new information is updated. The parameters in LSSVM are optimized based on an improved quantum-inspired version of Grey Wolf Optimizer (QGWO). The QGWO fortifies against the stagnation of the wolves at an optimal local point using three strategies including the quantum operator, non-linear convergence factor, and dynamic weighting. In this paper, the performance of QGWO was first demonstrated with that of other meta-heuristic (MH) algorithms in solving eight mathematical benchmark problems. Two time series of reservoir water availability in Zhoushan Islands, China, were then provided and analyzed to validate the performance of the advanced forecast approach in multi-step-ahead forecasts. The forecasts were compared with those obtained by the non-adaptive (NA) and non-recursive (NR) models. Results indicate that (1) QGWO displays superior performance on model parameters optimization than other comparative MH algorithms; (2) The proposed hybrid model in consideration of the nearest anterior feedback, as well as the adaptive mechanism, not only outperforms the two comparative models (NA, NR) but significantly enhances the accuracy of multi-step-ahead forecasts for non-stationary time series and low or high volume events, even as the forecast time horizon increases.

Forecasting U.S. Yield Curve Using the Dynamic Nelson–Siegel Model with Random Level Shift Parameters

In this paper, we develop a new model based on the classical dynamic Nelson-Siegel model by introducing random level shift (RLS) parameters. The built-in RLS can capture cyclical fluctuations in interest rates and structural breaks induced by technological progress, financial crisis, major monetary policy interventions, etc. In addition, the model can be used to forecast future structural breaks. We apply the model to fit and forecast daily U.S. Treasury yield curves and the model outperforms other widely used models. The empirical results show that the model not only has a better in-sample fit with residuals exhibiting less persistence but also has superior out-of-sample performance. Moreover, the model performs very well especially for short-term and long-term bonds, and the performance improves as the forecasting horizon increases.

Multi‐horizon accommodation demand forecasting: A New Zealand case study

AbstractThis paper contributes to the filling of two gaps in accommodation demand forecasting: (a) the limited number of studies on the use of modern machine learning techniques to identify the dynamics of accommodation demand; and (b) the lack of understanding of comparative forecasting performance of different modelling techniques at multiple forecast horizons. We show that, as the forecast horizon increases, the performance of machine learning is stable and robust. We also find that the long short‐term memory has particular advantages in long‐horizon forecasting and handling data with complex structure in New Zealand.

An observation regarding Hamilton’s recent criticisms of Kilian’s global real economic activity index

Hamilton (2019) offers several criticisms of the global real economic activity (REA) index suggested in Kilian (2009). This study deals with Hamilton (2019)’s arguments regarding the ability of the index to forecast commodity prices. Using in-sample population-level predictability tests, Hamilton (2019) argues that world industrial production index growth rate performs better than Kilian’s REA Index. We show that from an out-of-sample population-level predictability perspective, the REA index performs just as well if not better than world industrial production index. The Clark and West (2007) no population-level predictability null hypothesis is rejected at the same rate if not higher for the REA index, especially as the forecast horizon increases.

Predicting Intraday Prices in the Frontier Stock Market of Romania Using Machine Learning Algorithms

This paper investigates if forecasting models based on Machine Learning (ML) Algorithms are capable to predict intraday prices in the small, frontier stock market of Romania. The results show that this is indeed the case. Moreover, the prediction accuracy of the various models improves as the forecasting horizon increases. Overall, ML forecasting models are superior to the passive buy and hold strategy, as well as to a naïve strategy that always predicts the last known price action will continue. However, we also show that this superior predictive ability cannot be converted into “abnormal”, economically significant profits after considering transaction costs. This implies that intraday stock prices incorporate information within the accepted bounds of weak-form market efficiency, and cannot be “timed” even by sophisticated investors equipped with state of the art ML prediction models.

Deep Learning for Load Forecasting: Sequence to Sequence Recurrent Neural Networks With Attention

The biggest contributor to global warming is energy production and use. Moreover, a push for electrical vehicle and other economic developments are expected to further increase energy use. To combat these challenges, electrical load forecasting is essential as it supports energy production planning and scheduling, assists with budgeting, and helps identify saving opportunities. Machine learning approaches commonly used for energy forecasting such as feedforward neural networks and support vector regression encounter challenges with capturing time dependencies. Consequently, this paper proposes Sequence to Sequence Recurrent Neural Network (S2S RNN) with Attention for electrical load forecasting. The S2S architecture from language translation is adapted for load forecasting and a corresponding sample generation approach is designed. RNN enables capturing time dependencies present in the load data and S2S model further improves time modeling by combining two RNNs: encoder and decoder. The attention mechanism alleviates the burden of connecting encoder and decoder. The experiments evaluated attention mechanisms with different RNN cells (vanilla, LSTM, and GRU) and with varied time horizons. Results show that S2S with Bahdanau attention outperforms other models. Accuracy decreases as forecasting horizon increases; however, longer input sequences do not always increase accuracy.

Interpreting survey-based federal funds rate forecasts: how accurate are they in reflecting market expectations

What the market thinks is most likely to occur is not necessarily what the market expects to occur. This paper explains why most of the surveys of federal funds rate outlook deviate substantially from the true market expectation, especially as the forecast horizon increases. Surveys often ask participants for their forecast of the 'most likely outcome', which differs from the expected outcome. The latter has to take into account not only the most likely outcome but also those less likely to occur, that is, weighing all the possible outcomes by their probabilities. In a tightening (easing) cycle, the most likely outcome tends to be higher (lower) than the expected outcome, leading to a false impression that the fed will tighten (ease) more than what the market expects. It is only when the chances of rate hikes and rate cuts are roughly balanced that surveys reflect the true market expectation.

Short‐term forecasting of the US unemployment rate

AbstractThis paper aims to assess whether Google search data are useful when predicting the US unemployment rate among other more traditional predictor variables. A weekly Google index is derived from the keyword “unemployment” and is used in diffusion index variants along with the weekly number of initial claims and monthly estimated latent factors. The unemployment rate forecasts are generated using MIDAS regression models that take into account the actual frequencies of the predictor variables. The forecasts are made in real time, and the forecasts of the best forecasting models exceed, for the most part, the root mean squared forecast error of two benchmarks. However, as the forecasting horizon increases, the forecasting performance of the best diffusion index variants decreases over time, which suggests that the forecasting methods proposed in this paper are most useful in the short term.

Interpreting Survey-Based Federal Funds Rate Forecasts: How Accurate Are They in Reflecting Market Expectations?

What the market thinks is most likely to occur is not necessarily what the market expects to occur. This note explains why most of the surveys of federal funds rate outlook deviate substantially from the true market expectation, especially as the forecast horizon increases. Surveys often ask participants for their forecast of the “most likely outcome”, which differs from the expected outcome. The latter has to take into account not only the most likely outcome but also those less likely to occur, that is, weighing all the possible outcomes by their probabilities. In a tightening (easing) cycle, the most likely outcome tends to be higher (lower) than the expected outcome, leading to a false impression that the Fed will tighten (ease) more than what the market expects. For instance, the latest surveys suggest market participants believe there will most likely be another two to three rate hikes by the end of 2020, but in fact they expect the rate to stay basically unchanged. As the probability distribution gets more skewed to the left in progressive surveys, the “most likely outcome” forecast is increasingly subject to a greater downside risk. This, coupled with the relatively small increase in the probability at the high end of the forecasting range, means that market participants can progressively see the light at the end of the tunnel, that is, the end of the tightening cycle.